DESIGNING
EVALUATIONS
2012 Revision
Applied Research and Methods
January 2012
GAO-12-208G
United States Government Accountability Office
GAO
Page i GAO-12-208G
Preface 1
Chapter 1 The Importance of Evaluation Design
3
What Is a Program Evaluation? 3
Why Conduct an Evaluation? 4
Who Conducts Evaluations? 5
Why Spend Time on Design? 6
Five Key Steps to an Evaluation Design 7
For More Information 8
Chapter 2 Defining the Evaluation’s Scope
10
Clarify the Program’s Goals and Strategy 10
Develop Relevant and Useful Evaluation Questions 12
For More Information 16
Chapter 3 The Process of Selecting an Evaluation Design
18
Key Components of an Evaluation Design 18
An Iterative Process 20
Criteria for a Good Design 28
For More Information 29
Chapter 4 Designs for Assessing Program Implementation and Effectiveness
31
Typical Designs for Implementation Evaluations 31
Typical Designs for Outcome Evaluations 34
Typical Designs for Drawing Causal Inferences about Program
Impacts 39
Designs for Different Types of Programs 46
For More Information 48
Chapter 5 Approaches to Selected Methodological Challenges
50
Outcomes That Are Difficult to Measure 50
Complex Federal Programs and Initiatives 55
For More Information 61
Appendix I Evaluation Standards
64
“Yellow Book” of Government Auditing Standards 64
GAO’s Evaluation Synthesis 64
American Evaluation Association Guiding Principles for Evaluators 65
Contents
Page ii GAO-12-208G
Program Evaluation Standards, Joint Committee on Standards for
Educational Evaluation 65
Appendix II GAO Contact and Staff Acknowledgments
66
Other Papers in This Series
67
Tables
Table 1: Common Evaluation Questions Asked at Different Stages
of Program Development 15
Table 2: Common Designs for Implementation (or Process)
Evaluations 32
Table 3: Common Designs for Outcome Evaluations 36
Table 4: Common Designs for Drawing Causal Inferences about
Program Impacts 40
Table 5: Designs for Assessing Effectiveness of Different Types of
Programs 47
Figures
Figure 1: Sample Program Logic Model 11
Figure 2: Questions Guiding the Selection of Design Components 20
Page iii GAO-12-208G
Abbreviations
AEA American Evaluation Association
GAGAS generally accepted government auditing standards
GPRA Government Performance and Results Act of 1993
NSF National Science Foundation
OMB Office of Management and Budget
SAMHSA Substance Abuse and Mental Health Services
Administration
This is a work of the U.S. government and is not subject to copyright protection in the
United States. The published product may be reproduced and distributed in its entirety
without further permission from GAO. However, because this work may contain
copyrighted images or other material, permission from the copyright holder may be
necessary if you wish to reproduce this material separately.
Preface
Page 1 GAO-12-208G
GAO assists congressional decision makers in their deliberations by
furnishing them with analytical information on issues and options. Many
diverse methodologies are needed to develop sound and timely answers
to the questions the Congress asks. To provide GAO evaluators with
basic information about the more commonly used methodologies, GAO’s
policy guidance includes documents such as methodology transfer
papers and technical guides.
This methodology transfer paper addresses the logic of program
evaluation designs. It introduces key issues in planning evaluation studies
of federal programs to best meet decision makers’ needs while
accounting for the constraints evaluators face. It describes different types
of evaluations for answering varied questions about program
performance, the process of designing evaluation studies, and key issues
to consider toward ensuring overall study quality.
To improve federal program effectiveness, accountability and service
delivery, the Congress enacted the Government Performance and
Results Act of 1993 (GPRA), establishing a statutory framework for
performance management and accountability, including the requirement
that federal agencies set goals and report annually on progress towards
those goals and program evaluation findings. In response to this and
related management reforms, federal agencies have increased their
attention to conducting program evaluations. The GPRA Modernization
Act of 2010 raised the visibility of performance information by requiring
quarterly reviews of progress towards agency and governmentwide
priority goals. Designing Evaluations is a guide to successfully completing
evaluation design tasks. It should help GAO evaluators—and others
interested in assessing federal programs and policies—plan useful
evaluations and become educated consumers of evaluations.
Designing Evaluations is one of a series of papers whose purpose is to
provide guides to various aspects of audit and evaluation methodology
and indicate where more detailed information is available. It is based on
GAO studies and policy documents and program evaluation literature. To
ensure the guide’s competence and usefulness, drafts were reviewed by
selected GAO, federal and state agency evaluators, and evaluation
authors and practitioners from professional consulting firms. This paper
updates a 1991 version issued by GAO’s prior Program Evaluation and
Methodology Division. It supersedes that earlier version and incorporates
changes in federal program evaluation and performance measurement
since GPRA was implemented.
Preface
Designing Evaluations
Chapter 1: The Importance of Evaluation
Design
Page 3 GAO-12-208G
A program evaluation is a systematic study using research methods to
collect and analyze data to assess how well a program is working and
why. Evaluations answer specific questions about program performance
and may focus on assessing program operations or results. Evaluation
results may be used to assess a program’s effectiveness, identify how to
improve performance, or guide resource allocation.
There is no standard government definition of “program.” A program can
be defined in various ways for budgeting and policy-making purposes.
Whether a program is defined as an activity, project, function, or policy, it
must have an identifiable purpose or set of objectives if an evaluator is to
assess how well the purpose or objectives are met. Evaluations may also
assess whether a program had unintended (perhaps undesirable)
outcomes. An evaluation can assess an entire program or focus on an
initiative within a program. Although evaluation of a federal program
typically examines a broader range of activities than a single project,
agencies may evaluate individual projects to seek to identify effective
practices or interventions.
Program evaluation is closely related to performance measurement and
reporting. Performance measurement is the systematic ongoing
monitoring and reporting of program accomplishments, particularly
progress toward preestablished goals or standards. Performance
measures or indicators may address program staffing and resources (or
inputs), the type or level of program activities conducted (or process), the
direct products or services delivered by a program (or outputs), or the
results of those products and services (or outcomes) (GAO 2011).
A program evaluation analyzes performance measures to assess the
achievement of performance objectives but typically examines those
achievements in the context of other aspects of program performance or
in the context in which the program operates. Program evaluations may
analyze relationships between program settings and services to learn
how to improve program performance or to ascertain whether program
activities have resulted in the desired benefits for program participants or
the general public. Some evaluations attempt to isolate the causal
impacts of programs from other influences on outcomes, whereas
performance measurement typically does not. Evaluations have been
used to supplement performance reporting by measuring results that are
too difficult or expensive to assess annually or by exploring why
performance goals were not met. (For examples, see GAO 2000.)
Chapter 1: The Importance of Evaluation
Design
What Is a Program
Evaluation?
Chapter 1: The Importance of Evaluation
Design
Page 4 GAO-12-208G
Federal program evaluation studies are typically requested or initiated to
provide external accountability for the use of public resources (for
example, to determine the “value added” by the expenditure of those
resources) or to learn how to improve performance—or both. Evaluation
can play a key role in strategic planning and in program management,
providing feedback on both program design and execution.
Evaluations can be designed to answer a range of questions about
programs to assist decision-making by program managers and
policymakers. GAO evaluations are typically requested by congressional
committees to support their oversight of executive branch activities. A
committee might want to know whether agency managers are targeting
program funds to areas of greatest need or whether the program as
designed is, indeed, effective in resolving a problem or filling a need. The
Congress might use this information to reallocate resources for a more
effective use of funds or to revise the program’s design.
The Congress also directly requests agencies to report on program
activities and results. For example, legislative changes to a program
might be accompanied by a mandate that the agency report by a specific
date in the future on the effectiveness of those changes. Agencies may
choose to design an evaluation to collect new data if they are unable to
satisfy the request from available administrative data or performance
reporting systems. They may also evaluate pilot or demonstration projects
to inform the design of a new program.
GPRA performance reporting requirements were designed to provide
both congressional and executive decision makers with more objective
information on the relative effectiveness and efficiency of federal
programs and spending. However, due to the influence of other factors,
measures of program outcomes alone may provide limited information on
a program’s effectiveness. GPRA encourages federal agencies to
conduct evaluations by requiring agencies to (1) include a schedule of
future program evaluations in their strategic plans, (2) summarize their
evaluations’ findings when reporting annually on the achievement of their
performance goals, and (3) explain why a goal was not met. Federal
agencies have initiated evaluation studies to complement performance
measures by (1) assessing outcomes that are not available on a routine
or timely basis, (2) explaining the reasons for observed performance, or
(3) isolating the program’s impact or contribution to its outcome goals
(GAO 2000).
Why Conduct an
Evaluation?
Chapter 1: The Importance of Evaluation
Design
Page 5 GAO-12-208G
Since 2002, the Office of Management and Budget (OMB) under the
administrations of both Presidents Bush and Obama has set the
expectation that agencies should conduct program evaluations. Initial
OMB efforts to use agency performance reporting in decision making
were frustrated by the limited quantity and quality of information on results
(GAO 2005). Although federal program performance reporting improved,
in 2009 OMB initiated a plan to strengthen federal program evaluation,
noting that many important programs lacked evaluations and some
evaluations had not informed decision making (OMB 2009).
A federal program office or an agency research, policy or evaluation office
may conduct studies internally, or they may be conducted externally by
an independent consulting firm, research institute, or independent
oversight agency such as GAO or an agency’s Inspector General. The
choice may be based on where expertise and resources are available or
on how important the evaluator’s independence from program
management is to the credibility of the report. The choice may also
depend on how important the evaluator’s understanding of the program is
to the agency’s willingness to accept and act on the evaluation’s findings.
For example, evaluations aimed at identifying program improvement may
be conducted by a program office or an agency unit that specializes in
program analysis and evaluation. Professional evaluators typically have
advanced training in a variety of social science research methods.
Depending on the nature of the program and the evaluation questions,
the evaluation team might also require members with specialized subject
area expertise, such as labor economics. If agency staff do not have
specialized expertise or if the evaluation requires labor-intensive data
collection, the agency might contract with an independent consultant or
firm to obtain the required resources. (For more information, see U.S.
Department of Health and Human Services 2010.)
In contrast, evaluations conducted to provide an independent assessment
of a program’s strengths and weaknesses should be conducted by a team
independent of program management. Evaluations purchased by
agencies from professional evaluation firms can often be considered
independent. Conditions for establishing an evaluator’s independence
include having control over the scope, methods, and criteria of the review;
full access to agency data; and control over the findings, conclusions, and
recommendations.
Who Conducts
Evaluations?
Chapter 1: The Importance of Evaluation
Design
Page 6 GAO-12-208G
Evaluators have two basic reasons for taking the time to systematically
plan an evaluation: (1) to enhance its quality, credibility, and usefulness
and (2) to use their time and resources effectively.
A systematic approach to designing evaluations takes into account the
questions guiding the study, the constraints evaluators face in studying
the program, and the information needs of the intended users. After
exploring program and data issues, the initial evaluation question may
need to be revised to ensure it is both appropriate and feasible. Since the
rise in agency performance reporting, an enormous amount of program
information is available and there are myriad ways to analyze it. By
selecting the most appropriate measures carefully and giving attention to
the most accurate and reliable ways to collect data on them, evaluators
ensure the relevance of the analysis and blunt potential criticisms in
advance. Choosing well-regarded criteria against which to make
comparisons can lead to strong, defensible conclusions. Carefully
thinking through data and analysis choices in advance can enhance the
quality, credibility, and usefulness of an evaluation by increasing the
strength and specificity of the findings and recommendations. Focusing
the evaluation design on answering the questions being asked also will
likely improve the usefulness of the product to the intended users.
Giving careful attention to evaluation design choices also saves time and
resources. Collecting data through interviews, observation, or analysis of
records, and ensuring the quality of those data, can be costly and time
consuming for the evaluator as well as those subject to the evaluation.
Evaluators should aim to select the least burdensome way to obtain the
information necessary to address the evaluation question. When initiated
to inform decisions, an evaluation’s timeliness is especially important to
its usefulness. Evaluation design also involves considering whether a
credible evaluation can be conducted in the time and resources available
and, if not, what alternative information could be provided.
Developing a written evaluation design helps evaluators agree on and
communicate a clear plan of action to the project team and its advisers,
requestors, and other stakeholders, and it guides and coordinates the
project team’s activities as the evaluation proceeds. In addition, a written
plan justifying design decisions facilitates documentation of decisions and
procedures in the final report.
Why Spend Time on
Design?
Chapter 1: The Importance of Evaluation
Design
Page 7 GAO-12-208G
Evaluations are studies tailored to answer specific questions about how
well (or whether) a program is working. To ensure that the resulting
information and analyses meet decision maker’s needs, it is particularly
useful to isolate the tasks and choices involved in putting together a good
evaluation design. We propose that the following five steps be completed
before significant data are collected. These steps give structure to the
rest of this publication:
1. Clarify understanding of the program’s goals and strategy.
2. Develop relevant and useful evaluation questions.
3. Select an appropriate evaluation approach or design for each
evaluation question.
4. Identify data sources and collection procedures to obtain relevant,
credible information.
5. Develop plans to analyze the data in ways that allow valid conclusions
to be drawn from the evaluation questions.
The chapters in this paper discuss the iterative process of identifying
questions important to program stakeholders and exploring data options
(chapters 2 and 3) and the variety of research designs and approaches
that the evaluator can choose to yield credible, timely answers within
resource constraints (chapters 4 and 5). Completing an evaluation will, of
course, entail careful data collection and analysis, drawing conclusions
against the evaluation criteria selected, and reporting the findings,
conclusions, and recommendations, if any. Numerous textbooks on
research methods are adequate guides to ensuring valid and reliable data
collection and analysis (for example, Rossi et al. 2004, Wholey et al.
2010). GAO analysts are also urged to consult their design and
methodology specialists as well as the technical guides available on
GAO’s Intranet.
How evaluation results are communicated can dramatically affect how
they are used. Generally, evaluators should discuss preferred reporting
options with the evaluation’s requesters to ensure that their expectations
are met and prepare a variety of reporting formats (for example,
publications and briefings) to meet the needs of the varied audiences that
are expected to be interested in the evaluation’s results.
Five Key Steps to an
Evaluation Design
Chapter 1: The Importance of Evaluation
Design
Page 8 GAO-12-208G
GAO. 2011. Performance Measurement and Evaluation: Definitions and
Relationships, GAO-11-646SP. Washington, D.C. May.
GAO. 1998. Program Evaluation: Agencies Challenged by New Demand
for Information on Program Results, GAO/GGD-98-53. Washington, D.C.
Apr. 24.
GAO. 2005. Program Evaluation: OMB’s PART Reviews Increased
Agencies’ Attention to Improving Evidence of Program Results,
GAO-06-67. Washington, D.C. Oct. 28.
GAO. 2000. Program Evaluation: Studies Helped Agencies Measure or
Explain Program Performance, GAO/GGD-00-204. Washington, D.C.
Sept. 29.
American Evaluation Association. 2010. An Evaluation Roadmap for a
More Effective Government. www.eval.org/EPTF.asp
Bernholz, Eric, and others. 2006. Evaluation Dialogue Between OMB
Staff and Federal Evaluators: Digging a Bit Deeper into Evaluation
Science. Washington, D.C. July.
http://www.fedeval.net/docs/omb2006briefing.pdf
OMB (U. S. Office of Management and Budget). 2009. Increased
Emphasis on Program Evaluations, M-10-01, Memorandum for the Heads
of Executive Departments and Agencies. Washington, D.C.The White
House, Oct. 7.
Rossi, Peter H., Mark W. Lipsey, and Howard E. Freeman. 2004.
Evaluation: A Systematic Approach, 7th ed. Thousand Oaks, Calif.: Sage.
U.S. Department of Health and Human Services, Administration for
Children and Families, Office of Planning, Research and Evaluation.
2010. The Program Manager’s Guide to Evaluation, 2nd ed. Washington,
D.C. http://www.acf.hhs.gov/programs/opre/other_resrch/pm_guide_eval/
For More Information
GAO documents
Other resources
Chapter 1: The Importance of Evaluation
Design
Page 9 GAO-12-208G
Wholey, Joseph S., Harry P. Hatry, and Kathryn E. Newcomer. 2010.
Handbook of Practical Program Evaluation, 3rd ed. San Francisco, Calif.:
Jossey-Bass.
Chapter 2: Defining the Evaluation’s Scope
Page 10 GAO-12-208G
Because an evaluation can take any number of directions, the first steps
in its design aim to define its purpose and scope—to establish what
questions it will and will not address. The evaluation’s scope is tied to its
research questions and defines the subject matter it will assess, such as
a program or aspect of a program, and the time periods and locations that
will be included. To ensure the evaluation’s credibility and relevance to its
intended users, the evaluator must develop a clear understanding of the
program’s purpose and goals and develop researchable evaluation
questions that are feasible, appropriate to the program and that address
the intended users’ needs.
For some but not all federal programs, the authorizing legislation and
implementing regulations outline the program’s purpose, scope, and
objectives; the need it was intended to address; and who it is intended to
benefit. The evaluator should review the policy literature and consult
agency officials and other stakeholders to learn how they perceive the
program’s purpose and goals, the activities and organizations involved,
and the changes in scope or goals that may have occurred.
1
Evaluators use program logic models—flow diagrams that describe a
program’s components and desired results—to explain the strategy—or
logic—by which the program is expected to achieve its goals. By
specifying a theory of program expectations at each step, a logic model or
other representation can help evaluators articulate the assumptions and
expectations of program managers and stakeholders. In turn, by
specifying expectations, a model can help evaluators define measures of
the program’s performance and progress toward its ultimate goals. (For
examples, see GAO 2002.)
It is also
important to identify the program’s stage of maturity. Is the program still
under development, adapting to conditions on the ground, or is it a
complete system of activities purposefully directed at achieving agreed-on
goals and objectives? A program’s maturity affects the evaluator’s ability
to describe its strategy and anticipate likely evaluation questions.
At a minimum, a program logic model should outline the program’s inputs,
activities or processes, outputs, and both short-term and long-term
1
Program stakeholders are those individuals or groups with a significant interest in how
well the program functions, for example, decision makers, funders, administrators and
staff, and clients or intended beneficiaries.
Chapter 2: Defining the Evaluation’s Scope
Clarify the Program’s
Goals and Strategy
Chapter 2: Defining the Evaluation’s Scope
Page 11 GAO-12-208G
outcomes—that is, the ultimate social, environmental, or other benefits
envisioned. Including short-term and intermediate outcomes helps identify
precursors that may be more readily measured than ultimate benefits,
which may take years to achieve. It is also important to include any
external factors believed to have an important influence on—either to
hinder or facilitate—program inputs, operations, or achievement of
intended results. External factors can include the job market or other
federal or nonfederal activities aimed at the same outcomes. (Figure 1 is
a generic logic model developed for agricultural extension programs;
more complex models may describe multiple paths or perspectives.)
Figure 1: Sample Program Logic Model
A variety of formats can usefully assist in defining the evaluation’s scope;
the key is to develop a clear understanding of the nature of the program,
the context in which it operates, and the policy issues involved. A logic
model can be helpful as a:
Chapter 2: Defining the Evaluation’s Scope
Page 12 GAO-12-208G
• program planning tool: (reading from right to left) depicting the
implications for program design of previous research on the key
factors influencing achievement of the desired benefits;
• communication tool: encouraging shared understanding and
expectations among policy makers and program managers and
obtaining the support and cooperation of program partners;
• program implementation tool: mapping what activities should occur at
various times and which groups should be involved; and
• evaluation tool: helping to define performance measures and
formulate evaluation questions.
In describing a program’s goals and strategies, it is important to consult a
variety of sources—legislative history, program staff and materials, prior
research on the program, public media, congressional staff—to uncover
(if not resolve) any differences in expectations and concerns program
stakeholders have. It is also important to understand the program’s policy
context, why it was initiated, whether circumstances have changed
importantly since its inception, and what the current policy concerns are.
In the absence of clearly established definitions of the intervention or its
desired outcomes, the evaluator will need to discuss these issues with the
requestor and may need to explore, as part of the evaluation, how the
program and its goals have been operationally defined (see the
discussion of flexible grant programs in chapter 5).
Evaluation questions are constructed so that the issues and concerns of a
program’s stakeholders about program performance can be articulated
and to focus the evaluation to help ensure that its findings are useful
(GAO 2004). It is important to work with the evaluation requester to
formulate the right question to ensure that the completed evaluation will
meet his or her information needs. Care should be taken at this step
because evaluation questions frame the scope of the assessment and
drive the evaluation design—the selection of data to collect and
comparisons to make.
Program managers and policy makers may request information about
program performance to help them make diverse program management,
design, and budgeting decisions. Depending on the program’s history and
current policy context, the purpose for conducting an evaluation may be
Develop Relevant and
Useful Evaluation
Questions
Chapter 2: Defining the Evaluation’s Scope
Page 13 GAO-12-208G
to assist program improvement or to provide accountability, or both. More
specifically, evaluations may be conducted to
• ascertain the program’s progress in implementing key provisions,
• assess the extent of the program’s effectiveness in achieving desired
outcomes,
• identify effective practices for achieving desired results,
• identify opportunities to improve program performance,
• ascertain the success of corrective actions,
• guide resource allocation within a program, or
• support program budget requests.
These purposes imply different focuses—on the program as a whole or
just a component—as well as different evaluation questions and, thus,
designs. For example, if the purpose of the evaluation is to guide program
resource allocation, then the evaluation question might be tailored to
identify which program participants are in greatest need of services, or
which program activities are most effective in achieving the desired
results. To draw valid conclusions on which practices are most effective in
achieving the desired results, the evaluation might examine a few
carefully chosen sites in order to directly compare the effects of
alternative practices on the same outcomes, under highly comparable
conditions. (For further discussion see chapter 4 and GAO 2000.)
To be researchable, evaluation questions should be clear and specific
and use terms that can be readily defined and measured, and meet the
requester’s needs, so that the study’s scope and purpose are readily
understood and feasible. Evaluation questions should also be objective,
fair, and politically neutral; the phrasing of a question should not presume
to know the answer in advance.
Congressional requests for evaluations often begin with a very broad
concern, so discussion may be necessary to determine the requester’s
priorities and develop clearly defined researchable questions. Moreover,
while potentially hundreds of questions could be asked about a program,
limitations on evaluation resources and time require focusing the study on
Clarify the Issue
Chapter 2: Defining the Evaluation’s Scope
Page 14 GAO-12-208G
the most important questions that can be feasibly addressed. The
evaluator can use the program’s logic model to organize the discussion
systematically to learn whether the requester’s concerns focus on how
the program is operating or whether it is achieving its intended results or
producing unintended effects (either positive or negative). It is also
important to ensure that the evaluation question is well-matched to the
program’s purpose and strategies. For example, if a program is targeted
to meet the housing needs of low-income residents, then it would be
inappropriate to judge its effectiveness by whether the housing needs of
all residents were met.
It is important to learn whether the requester has a specific set of criteria
or expectations in mind to judge the program against and whether
questions pertain to the entire program or just certain components. A
general request to “assess a program’s effectiveness” should be clarified
and rephrased as a more specific question that ensures a common
understanding of the program’s desired outcomes, such as, “Has the
program led to increased access to health care for low-income
residents?” or “Has it led to lower incidence of health problems for those
residents?” It is also important to distinguish questions about the overall
effectiveness of a nationwide program from those limited to a few sites
that warrant study because they are especially promising or problematic.
The difference is extremely important for evaluation scope and design,
and attention to the difference allows the evaluator to help make the study
useful to the requester.
Although the feasibility of the evaluation questions will continue to be
assessed during the design phase, an evaluator should gain agreement
on these questions before completing the design of the evaluation. If
program stakeholders perceive the questions as objective and reflecting
their key concerns, they will be more likely to find the evaluation results
credible and persuasive and act on them.
Different questions tend to be asked at different stages of program
maturity and often reflect whether the purpose of the study is to assist
program improvement or provide accountability. Three types of evaluation
are defined by whether the focus is on the program’s operations or
outcomes, or on the program’s causal link to the observed results. Of
course, a single study may use different approaches to address multiple
questions. (See table 1.)
Ensure That Questions Are
Appropriate to the
Program’s Stage of
Maturity
Chapter 2: Defining the Evaluation’s Scope
Page 15 GAO-12-208G
Table 1: Common Evaluation Questions Asked at Different Stages of Program Development
Program stage Common evaluation questions Type of evaluation
Early stage of program or
new initiative within a
program
•
Is the program being delivered as intended to the targeted
recipients?
• Have any feasibility or management problems emerged?
• What progress has been made in implementing changes or new
provisions?
Process monitoring or
process evaluation
Mature, stable program with
well-defined program model
•
Are desired program outcomes obtained?
• What, if any, unintended side effects did the program produce?
• Do outcomes differ across program approaches, components,
providers, or client subgroups?
Outcome monitoring or
outcome evaluation
•
Are program resources being used efficiently?
• Why is a program no longer obtaining the desired level of
outcomes?
Process evaluation
•
Did the program cause the desired impact?
• Is one approach more effective than another in obtaining the desired
outcomes?
Net impact evaluation
Source Adapted from Bernholz et al 2006.
In the early stages of a new program or initiative within a program,
evaluation questions tend to focus on program process—on how well
authorized activities are carried out and reach intended recipients. Staff
need to be hired and trained, regulations written, buildings leased,
materials designed or purchased, participants identified and enrolled.
Program managers generally look for quick feedback on whether action is
needed to help get the program up and running as intended. Evaluation
studies designed to address the quality or efficiency of program
operations or their fidelity to program design are frequently called process
or implementation evaluations. Over time, some of the measures used to
evaluate program implementation may be institutionalized into an ongoing
program performance monitoring and reporting system. A process
evaluation can be an important companion to an outcome or impact
evaluation by describing the program as actually experienced.
Once assured that the program is operating as planned, one may ask
whether it is yielding the desired benefits or improvement in outcomes.
Outcome evaluations assess the extent to which a program achieves its
outcome-oriented objectives or other important outcomes. Naturally, if the
program has not had sufficient time to get its operations in place, then it is
unlikely to have produced the desired benefits. Depending on the nature
of the program, this shake-out period might take a few months, a year, or
perhaps even longer. In agreeing on an evaluation question, it is also
important to consider whether sufficient time will have passed to observe
Process Evaluations
Outcome Evaluations
Chapter 2: Defining the Evaluation’s Scope
Page 16 GAO-12-208G
longer-term outcomes. For example, it might take a study 3 or more years
to observe whether a program for high school students led to greater
success in college.
Where a program’s desired outcomes are known to also be influenced
appreciably by factors outside the program, such as the labor market, the
outcomes that are actually observed represent a combination of program
effects and the effects of those external factors. In this case, questions
about program effectiveness become more sophisticated and the
evaluation design should attempt to identify the extent to which the
program caused or contributed to those observed changes. Impact
evaluation is a form of outcome evaluation that assesses the net effect of
a program (or its true effectiveness) by comparing the observed
outcomes to an estimate of what would have happened in the absence of
the program. While outcome measures can be incorporated into ongoing
performance monitoring systems, evaluation studies are usually required
to assess program net impacts.
GAO. 2004. GAO’s Congressional Protocols, GAO-04-310G.
Washington, D.C.: July 16.
GAO. 2000. Managing for Results: Views on Ensuring the Usefulness of
Agency Performance Information to Congress, GAO/GGD-00-35.
Washington, D.C.: Jan. 26.
GAO. 2002. Program Evaluation: Strategies for Assessing How
Information Dissemination Contributes to Agency Goals, GAO-02-923.
Washington, D.C. Sept. 30.
Bernholz, Eric, and others. 2006. Evaluation Dialogue Between OMB
Staff and Federal Evaluators: Digging a Bit Deeper into Evaluation
Science. Washington, D.C.: July.
http://www.fedeval.net/docs/omb2006briefing.pdf
Rossi, Peter H., Mark W. Lipsey, and Howard E. Freeman. 2004.
Evaluation: A Systematic Approach, 7th ed. Thousand Oaks, Calif.: Sage.
Net Impact Evaluations
For More Information
GAO documents
Other resources
Chapter 2: Defining the Evaluation’s Scope
Page 17 GAO-12-208G
University of Wisconsin–Extension, Program Development and
Evaluation. www.uwex.edu/ces/pdande/evaluation/evallogicmodel.html
U.S. Department of Health and Human Services, Administration for
Children and Families, Office of Planning, Research and Evaluation.
2010. The Program Manager’s Guide to Evaluation, 2
nd
www.acf.hhs.gov/programs/opre/other_resrch/pm_guide_eval/
ed. Washington,
D.C.
Wholey, Joseph S., Harry P. Hatry, and Kathryn E. Newcomer. 2010.
Handbook of Practical Program Evaluation, 3rd ed. San Francisco:
Jossey-Bass.
Chapter 3: The Process of Selecting an
Evaluation Design
Page 18 GAO-12-208G
Once evaluation questions have been formulated, the next step is to
develop an evaluation design—to select appropriate measures and
comparisons that will permit drawing valid conclusions on those
questions. In the design process, the evaluator explores the variety of
options available for collecting and analyzing information and chooses
alternatives that will best address the evaluation objectives within
available resources. Selecting an appropriate and feasible design,
however, is an iterative process and may result in the need to revise the
evaluation questions.
An evaluation design documents the activities best able to provide
credible evidence on the evaluation questions within the time and
resources available and the logical basis for drawing strong conclusions
on those questions. The basic components of an evaluation design
include the following:
• the evaluation questions, objectives, and scope;
• information sources and measures, or what information is needed;
• data collection methods, including any sampling procedures, or how
information or evidence will be obtained;
• an analysis plan, including evaluative criteria or comparisons, or how
or on what basis program performance will be judged or evaluated;
• an assessment of study limitations.
Clearly articulating the evaluation design and its rationale in advance aids
in discussing these choices with the requester and other stakeholders.
Documenting the study’s decisions and assumptions helps manage the
study and assists report writing and interpreting results.
GAO evaluators outline the components of the evaluation design, as well
as the limitations of those choices, in a standard tool called a design
matrix. GAO evaluators are expected to complete a design matrix for
each significant project to document their decisions and summarize the
key issues in the evaluation design. All staff having significant
involvement in or oversight of the work meet to discuss this plan and
reach agreement on whether it can credibly answer the evaluation
questions.
Chapter 3: The Process of Selecting an
Evaluation Design
Key Components of
an Evaluation Design
GAO’s Design Matrix
Chapter 3: The Process of Selecting an
Evaluation Design
Page 19 GAO-12-208G
As a government oversight agency that conducts both audits and
evaluations, GAO also uses the design matrix to document and ensure
compliance with the government auditing fieldwork standards for
conducting performance audits (including program evaluations). The
fieldwork standards relate to planning, conducting, and documenting the
study. Government auditors are also expected to document in their plans
the implications of the agency’s internal controls, the results of previous
studies, and the reliability of agency databases for the evaluation’s scope
and objectives (GAO 2011).
The guidance for GAO’s design matrix is shown in figure 2 to
demonstrate the issues, design choices, and trade-offs that an evaluator
is expected to consider. Because GAO addresses a wide variety of
information requests in addition to program evaluations, the guidance is
fairly general but focuses on asking the evaluator to justify the design
components for each researchable question. Finally, the tool can help
stakeholders understand the logic of the evaluation.
Chapter 3: The Process of Selecting an
Evaluation Design
Page 20 GAO-12-208G
Figure 2: Questions Guiding the Selection of Design Components
Researchable
Question(s)
Information Required
and Source(s)
Scope and
Methodology Limitations
What This Analysis Will
Likely Allow GAO to
Say
What questions is the
team trying to answer?
Identify specific
questions that the team
must ask to address the
objectives in the
commitment letter and
job commitment report.
Ensure each major
evaluation question is
specific, objective,
neutral, measurable, and
doable. Ensure key terms
are defined.
Each major evaluation
question should be
addressed in a separate
row.
What information does
the team need to
address each
evaluation question?
Where will they get it?
Identify documents or
types of information that
the team must have.
Identify plans to address
internal controls and
compliance.
Identify plans to collect
documents that establish
the “criteria” to be used.
Identify plans to follow up
on known significant
findings and open
recommendations that
team found in obtaining
background information.
Identify sources of the
required information,
such as databases,
studies, subject area
experts, program
officials, models, etc.
How will the team
answer each evaluation
question?
Describe strategies for
collecting the required
information or data, such
as random sampling,
case studies, focus
groups, questionnaires,
benchmarking to best
practices, use of existing
data bases, etc.
Describe the planned
scope of each strategy,
including the timeframe,
locations to visit, and
sample sizes.
Describe the analytical
techniques to be used,
such as regression
analysis, cost benefit
analysis, sensitivity
analysis, modeling,
descriptive analysis,
content analysis, case
study summaries, etc.
What are the design’s
limitations and how will
it affect the product?
Cite any limitations as a
result of the information
required or the scope
and methodology, such
as:
—Questionable data
quality and/or reliability.
—Inability to access
certain types of data or
obtain data covering a
certain time frame.
—Security classification
or confidentiality
restrictions.
—Inability to generalize
or extrapolate findings to
the universe.
Be sure to address how
these limitations will
affect the product.
What are the expected
results of the work?
Describe what GAO can
likely say. Draw on
preliminary results for
illustrative purposes, if
helpful.
Ensure that the proposed
answer addresses the
evaluation question in
column one.
Source: GAO.
Designing an evaluation plan is iterative: evaluation objectives, scope,
and methodology are defined together because what determines them
often overlaps. Data limitations or new information about the program
may arise as work is conducted and have implications for the adequacy of
the original plans or the feasibility of answering the original questions. For
example, a review of existing studies of alternative program approaches
may uncover too few credible evaluations to support conclusions about
which approach is most effective. Thus, evaluators should consider the
need to make adjustments to the evaluation objectives, scope, and
methodology throughout the project.
An Iterative Process
Chapter 3: The Process of Selecting an
Evaluation Design
Page 21 GAO-12-208G
Nevertheless, the design phase of an evaluation is a period for examining
options for answering the evaluation questions and for considering which
options offer the strongest approach, given the time and resources
available. After reviewing materials about the program, evaluators should
develop and compare alternative designs and assess their strengths and
weaknesses. For example, in choosing between using program
administrative data or conducting a new survey of program officials, the
evaluator might consider whether 1) the new information collected
through a survey would justify the extra effort required, or 2) a high quality
survey can be conducted in the time available.
A key first step in designing an evaluation is to conduct a literature review
in order to understand the program’s history, related policies, and
knowledge base. A review of the relevant policy literature can help focus
evaluation questions on knowledge gaps, identify design and data
collection options used in the past, and provide important context for the
requester’s questions. An agency’s strategic plan and annual
performance reports can also provide useful information on available data
sources and measures and the efforts made to verify and validate those
data (GAO 1998).
Discussing evaluation plans with agency as well as congressional
stakeholders is important throughout the design process, since they have
a direct interest in and ability to act on the study’s findings. A principle of
good planning that helps ensure the transparency of our work is to notify
agency stakeholders of the evaluation’s scope and objectives at its outset
and discuss the expected terms of the work (GAO 2004). GAO evaluators
also coordinate their work with the Inspector General of the agency
whose program is being evaluated, and our sister congressional
agencies—the Congressional Budget Office and Congressional Research
Service—to avoid duplication, to leverage our resources, and to build a
mutual knowledge base. These meetings give evaluators opportunity to
learn about previous or ongoing studies and unfolding events that could
influence the design and use of the evaluation or necessitate modifying
the original evaluation question.
When a literature review reveals that several previous studies have
addressed the evaluation question, then the evaluator should consider
conducting a synthesis of their results before collecting new data. An
evaluation synthesis can answer questions about overall program
effectiveness or whether specific features of the program are working
Collect Background
Information
Consider Conducting an
Evaluation Synthesis
Chapter 3: The Process of Selecting an
Evaluation Design
Page 22 GAO-12-208G
especially well or especially poorly. Findings supported by a number of
soundly designed and executed studies add strength to the knowledge
base exceeding that of any single study, especially when the findings are
consistent across studies that used different methods. If, however, the
studies produced inconsistent findings, systematic analysis of the
circumstances and methods used across a number of soundly designed
and executed studies may provide clues to explain variations in program
performance (GAO 1992b). For example, differences between
communities in how they staff or execute a program or in their client
populations may explain differences in their effectiveness.
A variety of statistical approaches have been proposed for statistically
cumulating the results of several studies. A widely used procedure for
answering questions about program impacts is “meta-analysis,” which is a
way of analyzing “effect sizes” across several studies. Effect size is a
measure of the difference in outcome between a treatment group and a
comparison group. (For more information, see Lipsey and Wilson 2000.)
Depending on the program and study question, potential sources for
evidence on the evaluation question include program administrative
records, grantee reports, performance monitoring data, surveys of
program participants, and existing surveys of the national population or
private or public facilities. In addition, the evaluator may choose to
conduct independent observations or interviews with public officials,
program participants, or persons or organizations doing business with
public agencies.
In selecting sources of evidence to answer the evaluation question, the
evaluator must assess whether these sources will provide evidence that
is both sufficient and appropriate to support findings and conclusions on
the evaluation question. Sufficiency refers to the quantity of evidence—
whether it is enough to persuade a knowledgeable person that the
findings are reasonable. Appropriateness refers to the relevance, validity,
and reliability of the evidence in supporting the evaluation objectives. The
level of effort required to ensure that computer-processed data (such as
agency records) are sufficiently reliable for use will depend on the extent
to which the data will be used to support findings and conclusions and the
level of risk or sensitivity associated with the study. (See GAO 2009 for
more detailed guidance on testing the reliability of computer-processed
data.)
Assess the Relevance and
Quality of Available Data
Sources
Chapter 3: The Process of Selecting an
Evaluation Design
Page 23 GAO-12-208G
Measures are the concrete, observable events or conditions (or units of
evidence) that represent the aspects of program performance of interest.
Some evaluation questions may specify objective, quantifiable measures,
such as the number of families receiving program benefits, or qualitative
measures, such as the reasons for noncompliance. But often the
evaluator will need to select measures to represent a broader
characteristic, such as “service quality.” It is important to select measures
that clearly represent or are related to the performance they are trying to
assess. For example, a measure of the average processing time for tax
returns does not represent, and is not clearly related to, the goal of
increasing the accuracy of tax return processing. Measures are most
usefully selected in concert with the criteria that program performance will
be assessed against, so that agreement can be reached on the
sufficiency and appropriateness of the evidence for drawing conclusions
on those criteria.
Additional considerations for assessing the appropriateness of existing
databases include: whether certain subgroups of the population are well-
represented; whether converting data from its original format will require
excessive time or effort; and when examining multiple sites, whether
variation in data across sites precludes making reliable comparisons. No
data source is perfectly accurate and reliable; thus, evaluators often
consider using multiple measures or sources of data to triangulate toward
the truth. Concerns about biases in one data source—for example,
possible exaggerations in self reports of employment history— might be
countered by complementing that information with similar measures from
another source—for example, length of employment recorded in
administrative records.
No matter how data are collected, care should be taken to ensure that
data are sufficient and appropriate to support findings on the evaluation
question. Trained observers may inspect physical conditions, actions or
records to ascertain whether these met requirements or other kinds of
criteria, When collecting testimonial evidence through interviews or
surveys, the evaluator should consider whether the people serving as
data sources are sufficiently knowledgeable and whether their reports of
events or their opinions are likely to be candid and accurate. In addition,
careful attention to developing and pretesting questionnaire surveys and
other data collection instruments will help ensure that the data obtained
are sufficiently accurate for the purposes of the study. Where the
evaluator aims to aggregate and generalize from the results of a sample
survey, great importance is attached to collecting uniform data from every
Plan Original Data
Collection
Chapter 3: The Process of Selecting an
Evaluation Design
Page 24 GAO-12-208G
unit in the sample. Consequently, sample survey information is usually
acquired through structured interviews or self-administered
questionnaires. Most of the information is collected in close-ended form,
which means that the respondent chooses from responses offered in the
questionnaire or by the interviewer. Designing a consistent set of
responses into the data collection process helps establish the uniformity
of data across units in the sample. (For more on designing and
conducting surveys, see GAO 1991, Dillman 2007, Fowler 2009, or Willis
2005.)
A qualified survey specialist should be involved in designing and
executing questionnaire surveys that will be relied on for evidence on the
evaluation questions, whether the surveys are administered in person, by
telephone or mail, or over the Internet. Survey specialists can help ensure
that surveys are clearly understood, are quick and easy to complete, and
obtain the desired information. Subject matter experts should review the
survey to assess whether technical terms are used properly, respondents
are likely to have the desired information and will be motivated to
respond, and the questionnaire will provide a comprehensive, unbiased
assessment of the issues.
Federal executive agencies must adhere to guidance that OMB’s Office of
Information and Regulatory Affairs issues on policies and practices for
planning, implementing, and maintaining statistical activities, including
surveys used in program evaluations (OMB 2006). In addition, executive
branch agencies must submit certain proposals to collect information from
the public for OMB’s review and approval to ensure that they meet the
requirements of the Paperwork Reduction Act. GAO, as a legislative
branch agency, is not subject to these policies.
A potentially less costly alternative to conducting an original survey
(especially one with a large national sample) is to pay for additional
questions to be added to an ongoing national survey. This “piggy-back”
strategy is only useful, of course, if that survey samples the same
population needed for the evaluation. Another useful alternative data
collection approach is to link data from sample surveys to administrative
data systems, enabling the evaluator to obtain new information on, for
example, individuals, their neighborhoods, or their program participation.
(For more on record linkage and privacy protection procedures, see GAO
2001.)
Chapter 3: The Process of Selecting an
Evaluation Design
Page 25 GAO-12-208G
Evaluative criteria are the standards, measures, or expectations about
what should exist against which measures of actual performance are
compared and evaluated. Evaluators should select evaluative criteria that
are relevant, appropriate and sufficient to address the evaluation’s
objectives. Unlike financial or performance audits, the objectives of
program evaluations generally are not to assess a program’s or agency’s
compliance with legal requirements but to assess whether program
expectations have been met. The sources of those expectations can be
quite diverse. However, if the intended audience for the report—both the
study requesters and program managers—believes that the chosen
criteria and measures are appropriate, then the study’s findings are more
likely to be credible.
Depending on the circumstances of the program and the evaluation
questions, examples of possible criteria include
• purpose or goals prescribed by law or regulation,
• policies or procedures established by agency officials,
• professional standards or norms,
• expert opinions,
• prior period’s performance,
• performance of other entities or sectors used to benchmark
performance.
Some criteria designate a particular level as distinguishing acceptable
from unacceptable performance, such as in determinations of legal
compliance. Related evaluation questions ask whether a program’s
performance is “acceptable” or “meets expectations.” Other criteria have
no preestablished level designated as representing acceptable
performance but permit assessment of the extent to which expectations
are met. Thus, while the evaluation cannot typically ascertain whether a
program was “effective” per se, it can compare the performance of a
program across time and to the performance of other programs or
organizations to ascertain whether it is more or less effective than other
efforts to achieve a given objective.
To support objective assessment, criteria must be observable and
measurable events, actions, or characteristics that provide evidence that
Select Evaluative Criteria
Chapter 3: The Process of Selecting an
Evaluation Design
Page 26 GAO-12-208G
performance objectives have been met. Some legislation, evaluation
requests, or program designs provide broad concepts for performance
objectives, such as “a thorough process” or “family well-being,” that lack
clear assessment criteria. In such cases, the evaluator may need to gain
the agreement of study requesters and program managers to base
assessment criteria on measures and standards in the subject matter
literature.
In some cases, it makes sense to include all members of a population in a
study, especially where the population is small enough that it is feasible
within available resources and time periods to collect and analyze data on
the entire population (such as the 50 states)—called a certainty sample or
census. Many federal programs, however, cannot be studied by means of
a census and the evaluator must decide whether to collect data on a
probability or nonprobability sample.
In a probability sample (sometimes referred to as a statistical or random
sample), each unit in the population has a known, nonzero chance of
being selected. The results of a probability sample can usually be
generalized to the population from which the sample was taken. If the
objective is to report characteristics about a population, such as the
percentage of an agency’s officials who received certain training, or the
total dollar value of transactions in error in an agency’s system, then a
probability sample may be appropriate. A sampling specialist can help
identify how large a sample is needed to obtain precise estimates or
detect expected effects of a given size.
In a nonprobability sample, some units in the population have no chance,
or an unknown chance, of being selected. In nonprobability sampling, a
sample is selected from knowledge of the population’s characteristics or
from a subset of a population. Selecting locations to visit and identifying
officials to interview are part of many GAO studies, and these choices are
usually made using a nonprobability sampling approach. However, if it is
important to avoid the appearance of selection bias, locations or
interviewees can be selected using random sampling.
Deciding whether to use probability sampling is a key element of the
study design that flows from the scope of the researchable question. If the
question is, What progress has been made in implementing new program
provisions? then the implied study scope is program-wide and a
probability sample would be required to generalize conclusions drawn
from the locations observed to the program as a whole. In contrast, a
Select a Sample of
Observations
Chapter 3: The Process of Selecting an
Evaluation Design
Page 27 GAO-12-208G
question about why a program is no longer obtaining the desired level of
outcomes might be addressed by following up program locations that
have already been identified as not meeting the expected level of
outcomes—a purposive, nonprobability sample. A sampling specialist
should help select and design a sampling approach. (For more on
sampling, see GAO 1992a, Henry 1990, Lohr 2010, or Scheaffer et al.
2006.)
When engaging in primary (or original) data collection, it is important to
conduct a pretest or pilot study before beginning full-scale data collection.
The pilot study gives the evaluator an opportunity to refine the design and
test the availability, reliability, and appropriateness of proposed data.
Evaluators new to the program or proposing new data collection may find
that a limited exploration of the proposed design in a few sites can
provide a useful “reality check” on whether one’s assumptions hold true.
The pilot phase allows for a check on whether program operations, such
as client recruitment, and delivery of services occur as expected. Finding
that they do not may suggest a need to refocus the evaluation question to
ask why the program has been implemented so differently from what was
proposed. Testing the work at one or more sites allows the evaluator to
confirm that data are available, the form they take, and the means for
gathering them, including interview procedures. It also provides an
opportunity to assess whether the analysis methodology will be
appropriate.
Existing data sources should be closely examined for their suitability for
the planned analyses. For example, to support sophisticated statistical
analyses, data may be needed as actual dollars, days, or hours rather
than aggregated into a few wide ranges. To ensure the ability to reliably
assess change over time, the evaluator should check whether there have
been changes in data recording, coding, or storage procedures over the
period of interest.
Evaluators need to work with the stakeholders and acknowledge what the
study can and cannot address when making the project’s scope and
design final. The end of the design phase is an important milestone. It is
here that the evaluator must have a clear understanding of what has been
chosen, what has been omitted, what strengths and weaknesses have
been embedded in the design, what the customer’s needs are, how
usefully the design is likely to meet those needs, and whether the
constraints of time, cost, staff, location, and facilities have been
Pilot Test Data Collection
and Analysis Procedures
Assess Study Limitations
Chapter 3: The Process of Selecting an
Evaluation Design
Page 28 GAO-12-208G
adequately addressed. Evaluators must be explicit about the limitations of
the study. They should ask, How conclusive is the study likely to be, given
the design? How detailed are the data collection and data analysis plans?
What trade-offs were made in developing these plans?
GAO and other organizations have developed guidelines or standards to
help ensure the quality, credibility, and usefulness of evaluations. (See
appendix I and the guidance in GAO’s design matrix, figure 2, as an
example.) Some standards pertain specifically to the evaluator’s
organization (for example, whether a government auditor is independent),
the planning process (for example, whether stakeholders were
consulted), or reporting (for example, documenting assumptions and
procedures). While the underlying principles substantially overlap, the
evaluator will need to determine the relevance of each guideline to the
evaluator’s organizational affiliation and their specific evaluation’s scope
and purpose.
Strong evaluations employ methods of analysis that are appropriate to the
question; support the answer with sufficient and appropriate evidence;
document the assumptions, procedures, and modes of analysis; and rule
out competing explanations. Strong studies present questions clearly,
address them appropriately, and draw inferences commensurate with the
power of the design and the availability, validity, and reliability of the data.
Thus, a good evaluation design should
• be appropriate for the evaluation questions and context. The design
should address all key questions, clearly state any limitations in
scope, and be appropriate to the nature and significance of the
program or issue. For example, evaluations should not attempt to
measure outcomes before a program has been in place long enough
to be able to produce them.
• adequately address the evaluation question. The strength of the
design should match the precision, completeness, and
conclusiveness of the information needed to answer the questions
and meet the client’s needs. Criteria and measures should be
narrowly tailored, and comparisons should be selected to support
valid conclusions and rule out alternative explanations.
• fit available time and resources. Time and cost are constraints that
shape the scope of the evaluation questions and the range of
Criteria for a Good
Design
Chapter 3: The Process of Selecting an
Evaluation Design
Page 29 GAO-12-208G
activities that can help answer them. Producing information with an
understanding of the user’s timetable enhances its usefulness.
• rely on sufficient, credible data. No data collection and maintenance
process is free of error, but the data should be sufficiently free of bias
or other significant errors that could lead to inaccurate conclusions.
Measures should reflect the persons, activities, or conditions that the
program is expected to affect and should not be unduly influenced by
factors outside the program’s control.
GAO. 1992a. Using Statistical Sampling, revised, GAO/PEMD-10.1.6.
Washington, D.C. May.
Henry, Gary T. 1990. Practical Sampling. Thousand Oaks, Calif.: Sage.
Lohr, Sharon L. 2010. Sampling: Design and Analysis, 2
nd
Scheaffer, Richard L., William Mendenhall III, and R. Lyman Ott. 2006.
Elementary Survey Sampling, 6th ed. Cengage Learning.
ed.
Brooks/Cole, Cengage Learning.
Dillman, Don A. 2007. Mail and Internet Surveys: The Tailored Design
Method, 2nd ed. New York: Wiley.
Fowler, Floyd J., Jr. 2009. Survey Research Methods, 4
th
GAO. 1991. Using Structured Interviewing Techniques.
ed. Thousand
Oaks, Calif.: Sage.
GAO/PEMD-10.1.5. Washington, D.C. June.
Willis, Gordon B. 2005. Cognitive Interviewing: A Tool for Improving
Questionnaire Design. Thousand Oaks, Calif.: Sage.
American Evaluation Association. 2004. Guiding Principles for Evaluators.
July. www.eval.org/Publications/GuidingPrinciples.asp.
For More Information
On sampling approaches
On developing surveys and
questionnaires
On standards
Chapter 3: The Process of Selecting an
Evaluation Design
Page 30 GAO-12-208G
GAO. 2011. Government Auditing Standards: 2011 Internet Version.
Washington, D.C. August. http://www.gao.gov/govaud/iv2011gagas.pdf
GAO. 1992b. The Evaluation Synthesis, revised, GAO/PEMD-10.1.2.
Washington, D.C. March.
Yarbrough, Donald B., Lynn M. Shulha, Rodney K. Hopson, and Flora A.
Caruthers. 2011. The Program Evaluation Standards: A Guide for
Evaluators and Evaluation Users, 3rd ed. Thousand Oaks, Calif.: Sage.
GAO. 2009. Assessing the Reliability of Computer-Processed Data,
external version 1. GAO-09-680G. Washington, D.C. July.
GAO. 2004. GAO’s Agency Protocols, GAO-05-35G. Washington, D.C.
October.
GAO. 2001. Record Linkage and Privacy: Issues in Creating New Federal
Research and Statistical Information. GAO-01-126SP. Washington, D.C.
April.
GAO. 1998. The Results Act: An Evaluator’s Guide to Assessing Agency
Annual Performance Plans, version 1. GAO/GGD-10.1.20. Washington,
D.C. April.
Lipsey, Mark W., and David R. Wilson. 2000. Practical Meta-Analysis.
Thousand Oaks, Calif.: Sage.
OMB (U.S. Office of Management and Budget), Office of Information and
Regulatory Affairs. 2006. Standards and Guidelines for Statistical
Surveys. Washington, D.C. September.
http://www.whitehouse.gov/omb/inforeg_statpolicy#pr
Other resources
Chapter 4: Designs for Assessing Program
Implementation and Effectiveness
Page 31 GAO-12-208G
Program evaluation designs are tailored to the nature of the program and
the questions being asked. Thus, they can have an infinite variety of
forms as evaluators choose performance goals and measures and select
procedures for data collection and analysis. Nevertheless, individual
designs tend to be adaptations of a set of familiar evaluation
approaches—that is, evaluation questions and research methods for
answering them (Rossi et al. 2004). This chapter provides examples of
some typical evaluation approaches for implementation and effectiveness
questions and examples of designs specifically matched to program
structure. Chapter 5 provides examples of approaches to evaluating
programs where either the intervention or desired outcomes are not
clearly defined.
Implementation (or process) evaluations address questions about how
and to what extent activities have been implemented as intended and
whether they are targeted to appropriate populations or problems.
Implementation evaluations are very similar to performance monitoring in
assessing the quality and efficiency of program operations, service
delivery, and service use, except that they are conducted as separate
projects, not integrated into the program’s daily routine. Implementation
evaluations may be conducted to provide feedback to program managers,
accountability to program sponsors and the public, or insight into variation
in program outcomes. These evaluations may answer questions such as
• Are mandated or authorized activities being carried out?
• To what extent is the program reaching the intended population?
• Have feasibility or management problems emerged?
• Why is the program no longer achieving its expected outcomes?
Assessing how well a program is operating requires first identifying a
criterion against which a program’s performance is compared.
Alternatively, an assessment may compare performance across locations,
points in time, or subgroups of the population, to identify important
variations in performance. In contrast, an exploratory case study of
program processes and context may focus on exploring reasons why the
program is operating as it is. Table 2 provides examples of
implementation questions and designs used to address them.
Chapter 4: Designs for Assessing Program
Implementation and Effectiveness
Typical Designs for
Implementation
Evaluations
Chapter 4: Designs for Assessing Program
Implementation and Effectiveness
Page 32 GAO-12-208G
Table 2: Common Designs for Implementation (or Process) Evaluations
Evaluation question Design
Is the program being implemented as intended? Compare program activities to statute and regulations, program
logic model, professional standards, or stakeholder expectations
Have any feasibility or management problems emerged?
•
Compare program performance to quality, cost or efficiency
expectations
• Assess variation in quality or performance across settings,
providers, or subgroups of recipients
Why is the program not (or no longer) achieving expected
outcomes?
•
Analyze program and external factors correlated with
variation in program outcomes
• Interview key informants about possible explanations
• Conduct indepth analysis of critical cases
Source GAO.
Assessments of program implementation often compare program
performance—or what is—to a criterion established in advance—or what
should be. The evaluative criteria may be derived from the law,
regulations, a program logic model, administrative or professional
standards, research identifying the best practices of leading
organizations, or stakeholder expectations. Some criteria identify an
acceptable level of performance or performance standard by, for
example, defining authorized activities. In some areas, a program may
not be considered credible unless it meets well-established professional
standards. When criteria have no predetermined standard of acceptable
performance, the evaluator’s task is to measure the extent to which a
program meets its objectives. Measures of program performance may be
obtained from program records or may be specially collected for the
evaluation through interviews, observations, or systems testing. For
example,
• To assess the quality, objectivity, utility, and integrity of an agency’s
statistical program, an evaluator can compare its policies and
procedures for designing, collecting, processing, analyzing and
disseminating data with government guidelines for conducting
statistical surveys (OMB 2006).
• To evaluate the operational quality and efficiency of a program
providing financial assistance to individuals, an evaluator might
analyze administrative records that document the applications
received for program benefits and the actions taken on them.
Efficiency might be assessed by how promptly applications for
benefits were processed for a given level of staffing; quality might be
Assessing Quality or the
Progress of Program
Implementation
Chapter 4: Designs for Assessing Program
Implementation and Effectiveness
Page 33 GAO-12-208G
assessed by how accurately eligibility and benefits were determined
(GAO 2010). Standards of acceptable or desired performance might
be drawn from previous experience or the levels of quality assurance
achieved in other financial assistance programs.
• To evaluate a program’s success in serving a target population such
as low-income children, one might analyze program records to
compare the family incomes of current participants to the national
poverty level or to family income levels of recipients in previous years.
However, to address how well the program is reaching the population
eligible for the program, a better choice might be to compare
information from local program records with surveys of the income of
local residents to estimate the proportion of the local low-income
population that the program reached. To assess improvement in
program targeting, the evaluator could compare that program
coverage statistic over time. However, additional analysis would be
required to ascertain whether observed improvements in coverage
resulted from program improvements or changes in the neighborhood.
To identify program management or feasibility issues in federal programs,
it is often important to examine the nature and sources of variation in
program quality or performance across settings, providers, or population
subgroups. For example,
• To evaluate how well a new technical assistance program is
operating, an evaluator might review program records as well as
survey local program managers to learn whether any feasibility
problems had developed. Program records might address whether
guidance materials were issued and delivered in a timely manner or
whether workshops were held promptly and drew the attendance
expected. But an evaluator might also want to survey local managers
for their judgments on whether the guidance and training materials
were technically competent and relevant to their needs. Performance
standards might be drawn from program design and planning
materials, program technical standards, or previous experience with
needs for technical assistance.
Because of the cost of collecting and analyzing data on all program
participants or transactions, evaluators of federal programs frequently
collect data by surveying a nationally representative probability sample.
Sample surveys can also address questions about variation in service
delivery across geographic locations or types of providers.
Assessing Variation in
Implementation
Chapter 4: Designs for Assessing Program
Implementation and Effectiveness
Page 34 GAO-12-208G
In some circumstances, an evaluator may want to use case studies to
explore certain issues in more depth than can be done in more than a few
locations. In single case study evaluations, especially, much attention is
given to acquiring qualitative information that describes events and
conditions from several points of view. The structure imposed on the data
collection may range from the flexibility of ethnography or investigative
reporting to the highly structured interviews of sample surveys. (For more
on the evaluation insights to be gained from ethnography, see GAO
2003.) Case studies are often used to provide in-depth descriptive
information about how the program operates in the field. If the objective of
the case study is to describe aspects of an issue, provide context, or
illustrate findings developed from a more broadly applied survey, then
selecting a nongeneralizable sample of cases may be appropriate.
Case studies can also supplement survey or administrative data to
explore specific questions about program performance, such as
understanding variation in program performance across locations (for
example, rural versus urban settings), or to identify factors key to program
success or failure. The criteria used for selecting cases are critical to
one’s ability to apply their findings to the larger program. To heighten the
value of the information they provide, cases should be selected carefully
to represent particular conditions of interest (for example, sites with low
versus high levels of performance) and with certain hypotheses in mind.
However, most often, case studies will generate hypotheses rather than
answers to questions such as what factors influence program success.
(For more on case study methodology, see GAO 1990, Stake 1995, or
Yin 2009.) For example,
• To identify the causes of a sudden decline in control of an agricultural
pest, evaluators might conduct field observations in the localities most
affected to assess how well key components of the pest eradication
and control program were executed or whether some other factor
appeared to be responsible.
Outcome evaluations address questions about the extent to which the
program achieved its results-oriented objectives. This form of evaluation
focuses on examining outputs (goods and services delivered by a
program) and outcomes (the results of those products and services) but
may also assess program processes to understand how those outcomes
are produced. Outcome evaluations may address questions such as
Case Studies
Typical Designs for
Outcome Evaluations
Chapter 4: Designs for Assessing Program
Implementation and Effectiveness
Page 35 GAO-12-208G
• Is the program achieving its intended purposes or objectives?
• Has it had other important (unintended) side effects on issues of
stakeholder concern?
• Do outcomes differ across program approaches, components,
providers, or client subgroups?
• How does the program compare with other strategies for achieving
the same ends?
To appropriately assess program effectiveness, it is important, first, to
select outcome measures that clearly represent the nature of the
expected program benefit, cover key aspects of desired performance, and
are not unduly influenced by factors outside the program’s control. Next,
to allow causal inferences about program effects, the data collection and
analysis plan must establish a correlation between exposure to the
program and the desired benefit and must set a time-order relationship
such that program exposure precedes outcomes.
However, if the evaluators suspect that factors outside the program
appreciably influenced the observed outcomes, then they should not
present the findings of an outcome evaluation as representing the results
caused by the program. Instead, they should choose one of the net
impact designs discussed in the next section to attempt to isolate effects
attributable to the program. Ongoing monitoring of social conditions such
as a community’s health or employment status can provide valuable
feedback to program managers and the public about progress toward
program goals but may not directly reflect program performance. Table 3
provides examples of outcome-oriented evaluation questions and designs
used to address them.
Chapter 4: Designs for Assessing Program
Implementation and Effectiveness
Page 36 GAO-12-208G
Table 3: Common Designs for Outcome Evaluations
Evaluation question Design
Is the program achieving its desired outcomes or having other
important side effects?
•
Compare program performance to law and regulations,
program logic model, professional standards, or
stakeholder expectations
• Assess change in outcomes for participants before and
after exposure to the program
• Assess differences in outcomes between program
participants and nonparticipants
Do program outcomes differ across program components,
providers or recipients?
Assess variation in outcomes (or change in outcomes) across
approaches, settings, providers, or subgroups of recipients
Source GAO.
Like outcome monitoring, outcome evaluations often assess the benefits
of the program for participants or the broader public by comparing data on
program outcomes to a preestablished target value. The criterion could
be derived from law, regulation, or program design, while the target value
might be drawn from professional standards, stakeholder expectations, or
the levels observed previously in this or similar programs. This can help
ensure that target levels for accomplishments, compliance, or absence of
error are realistic. For example,
• To assess the immediate outcomes of instructional programs, an
evaluator could measure whether participants’ experienced short-term
changes in knowledge, attitudes, or skills at the end of their training
session. The evaluator might employ post-workshop surveys or
conduct observations during the workshops to document how well
participants understood and can use what was taught. Depending on
the topic, industry standards might provide a criterion of 80 percent or
90 percent accuracy, or demonstration of a set of critical skills, to
define program success. Although observational data may be
considered more accurate indicators of knowledge and skill gains than
self-report surveys, they can often be more resource-intensive to
collect and analyze.
In programs where there are quantitative measures of performance but
no established standard or target value, outcome evaluations at least may
rely on assessing change or differences in desired outputs and outcomes.
The level of the outcome of interest, such as client behavior or
environmental conditions, is compared with the level observed in the
absence of the program or intervention. This can be done by comparing
Assessing the Achievement
of Intended Outcomes
Assessing Change in
Outcomes
Chapter 4: Designs for Assessing Program
Implementation and Effectiveness
Page 37 GAO-12-208G
• the behavior of individuals before and after their exposure to a
program,
• environmental conditions before and after an intervention, or
• the outcomes for individuals who did and did not participate in the
program.
Of course, to conclude that any changes observed reflect program
effects, the evaluator must feel confident that those changes would not
have occurred on their own without the program, in response to some
nonprogram influences. For example,
• The accuracy and timeliness of severe weather forecasts—arguably
considered program outputs—can be compared to target levels of
performance through analysis of program records over time. However,
it is more problematic to attempt to assess the effectiveness of the
forecasting program through the amount of harm resulting from those
storms—what might be considered program outcomes. This is
because building construction and evacuation policies—external
factors to a weather forecasting program—are also expected to
greatly influence the amount of harm produced by a storm.
• To assess an industry’s compliance with specific workplace safety
regulations, an evaluator could conduct work-site observations or
review agency inspections records and employer injury and illness
reports. The evaluator might analyze changes in compliance and
safety levels at work sites after a regulation was enacted or compare
compliance and safety levels between employers who were or were
not provided assistance in complying with the regulations. Again,
however, to draw conclusions about the effectiveness or impact of the
regulation (or compliance assistance) in improving worker safety, the
evaluator needs to be able to rule out the influence of other possible
workplace changes, such as in technology, worker experience, or
other aspects of working conditions.
As in process evaluations, sample surveys can be used to collect
outcome data on probability samples in order to provide information about
the program as a whole. A cross-sectional survey, the simplest form of
sample survey, takes measurements at a point in time to describe events
or conditions. By providing information on the incidence of events or
distribution of conditions in relationship to a preselected standard or
target value, it can be used to assess program performance in either a
Chapter 4: Designs for Assessing Program
Implementation and Effectiveness
Page 38 GAO-12-208G
process or an outcome evaluation. Through repeated application, a cross-
sectional survey can measure change over time for the population as a
whole. A panel survey acquires information from the same sample units
at two or more points in time. Thus, a panel survey can provide less
variable measures of change in facts, attitudes, or opinions over time and
thus can support more directly comparative assessments of outcomes
than can a cross-sectional survey, although often at greater cost. Adding
the important element of time helps in drawing inferences with regard to
cause and effect.
Variation in outcomes across settings, providers or populations can be
the result of variation in program operations (such as level of
enforcement) or context (such as characteristics of client populations or
settings). Variation in outcomes associated with features under program
control, such as the characteristics of service providers or their activities,
may identify opportunities for managers to take action to improve
performance. However, additional information is usually needed to
understand why some providers are obtaining worse results than others—
for example, whether the staff lack needed skills or are ineffectively
managed. Variation associated with factors outside the control of the
program, such as neighborhood characteristics, can help explain program
results, but may not identify actions to improve program performance.
Thus, although analysis of surveys or performance reports can identify
factors correlated with variation in outcomes, follow-up studies or more
complex designs (see the next section) are needed to draw firm
conclusions about their likely causes.
Case studies are not usually used to assess program effectiveness
because their results cannot be generalized to the program as a whole
and because of the difficulty of distinguishing many possible causes of a
unique instance. However, in special circumstances, an outcome
evaluation may use a case study to examine a critical instance closely to
understand its cause or consequences. Often such a study is an
investigation of a specific problem event, such as a fatal accident or forest
fire. The potential causal factors can be numerous and complex, requiring
an in-depth examination to assess whether and which safety program
components were ineffective in preventing or responding to that event.
Critical incident studies are also discussed in chapter 5.
Assessing Variation in
Outcomes
Chapter 4: Designs for Assessing Program
Implementation and Effectiveness
Page 39 GAO-12-208G
Many desired outcomes of federal programs are influenced by external
factors, including other federal, state, and local programs and policies, as
well as economic or environmental conditions. Thus, the outcomes
observed typically reflect a combination of influences. To isolate the
program’s unique impacts, or contribution to those outcomes, an impact
study must be carefully designed to rule out plausible alternative
explanations for the results. Typical approaches to this problem include
• selection of targeted outcome measures,
• comparison group research designs,
• statistical analysis, and
• logical argument.
A well-articulated program logic model is quite valuable in planning an
impact evaluation. Clearly articulating the program’s strategy and
performance expectations aids the selection of appropriate performance
measures and data sources. Identifying the most important external
influences on desired program outcomes helps in developing research
designs that convincingly rule out the most plausible alternative
explanations for the observed results.
Impact evaluation research designs construct comparisons of what
happened after exposure to the program with an estimate of what would
have happened in the absence of the program in order to estimate the net
impact of the program. A number of methodologies are available to
estimate program impact, including experimental, quasi-experimental,
and nonexperimental designs. Conducting an impact evaluation of a
social intervention often requires the expenditure of significant resources
to collect and analyze data on program results and estimate what would
have happened in the absence of the program. Thus, impact evaluations
need not be conducted for all interventions but should be reserved for
when the effort and cost appear warranted: for an intervention that is
important, clearly defined, well-implemented, and being considered for
adoption elsewhere (GAO 2009). Table 4 provides examples of designs
commonly used to address net impact questions.
Typical Designs for
Drawing Causal
Inferences about
Program Impacts
Chapter 4: Designs for Assessing Program
Implementation and Effectiveness
Page 40 GAO-12-208G
Table 4: Common Designs for Drawing Causal Inferences about Program Impacts
Evaluation question Design
Is the program responsible for (effective in) achieving
improvements in desired outcomes?
•
Compare (change in) outcomes for a randomly assigned
treatment group and a nonparticipating control group
(randomized controlled experiment)
• Compare (change in) outcomes for program participants and
a comparison group closely matched to them on key
characteristics (comparison group quasi-experiment)
• Compare (change in) outcomes for participants before and
after the intervention, over multiple points in time with
statistical controls (single group quasi-experiment)
How does the effectiveness of the program approach compare
with other strategies for achieving the same outcomes?
•
Compare (change in) outcomes for groups randomly
assigned to different treatments (randomized controlled
experiment)
• Compare (change in) outcomes for comparison groups
closely matched on key characteristics (comparison group
quasi-experiment)
Source Adapted from Bernholz et al 2006.
The defining characteristic of an experimental design is that units of study
are randomly assigned either to a treatment (or intervention) group or to
one or more nonparticipating control (or comparison) groups. Random
assignment means that the assignment is made by chance, as in the flip
of a coin, in order to control for any systematic difference between the
groups that could account for a difference in their outcomes. A difference
in these groups’ subsequent outcomes is believed to represent the
program’s impact because, under random assignment, the factors that
influence outcomes other than the program itself should be evenly
distributed between the two groups; their effects tend to cancel one
another out in a comparison of the two groups’ outcomes. A true
experiment is seldom, if ever, feasible for GAO because evaluators must
have control over the process by which participants in a program are
assigned to it, and this control generally rests with the agency. However,
GAO does review experiments carried out by others.
Depending on how the program is administered, the unit of study might
be such entities as a person, classroom, neighborhood, or industrial plant.
More complex designs may involve two or more comparison groups that
receive different combinations of services or experience the program at
different levels of intensity. For example, patients might be randomly
assigned to drug therapy, dietary, or exercise interventions to treat high
blood pressure. For example,
Randomized Experiments
Chapter 4: Designs for Assessing Program
Implementation and Effectiveness
Page 41 GAO-12-208G
• To evaluate the effect of the provision of housing assistance and
employment support services on the capacity of low-income families
to obtain or retain employment, the Department of Housing and Urban
Development conducted a randomized experiment. In the sites
chosen for the evaluation, eligible families on the waiting list for
housing subsidies were randomly assigned either to an experimental
group, who received a voucher and the employment support services
bound to it, or to a control group, who did not receive a voucher or
services. Both groups have been tracked for several years to
determine the impact of the provision of rental assistance and
accompanying services on families’ employment, earnings, and
geographic mobility (Abt Associates and QED Group 2004).
Randomized experiments are best suited for assessing intervention or
program effectiveness when it is possible, ethical, and practical to
conduct and maintain random assignment to minimize the influence of
external factors on program outcomes. Some kinds of interventions are
not suitable for randomized assignment because the evaluator needs to
have control over who will be exposed to it, and that may not be possible.
Examples include interventions that use such techniques as public
service announcements broadcast on the radio, television, or Internet.
Random assignment is well suited for programs that are not universally
available to the entire eligible population, so that some people will be
denied access to services in any case, and a lottery is perceived as a fair
way to form a comparison group.
Thus, no comparison group design is possible to assess full program
impact where agencies are prohibited from withholding benefits from
individuals entitled to them (such as veterans’ benefits) or from selectively
applying a law to some people but not others. Random assignment is
often not accepted for testing interventions that prevent or mitigate harm
because it is considered unethical to impose negative events or elevated
risks of harm to test a remedy’s effectiveness. Instead, the evaluator must
wait for a hurricane or flood, for example, to learn if efforts to strengthen
buildings prevented serious damage. (For further discussion, see GAO
2009, Rossi et al. 2004, or Shadish et al. 2002.)
Field experiments are distinguished from laboratory experiments and
experimental simulations in that field experiments take place in much less
contrived, more naturalistic settings such as classrooms, hospitals, or
workplaces. Conducting an inquiry in the field gives reality to the
evaluation but often at the expense of some accuracy in the results. This
is because experiments conducted in field settings allow limited control
Limited Applicability of
Randomized Experiments
Difficulties in Conducting Field
Experiments
Chapter 4: Designs for Assessing Program
Implementation and Effectiveness
Page 42 GAO-12-208G
over both program implementation and external factors that may influence
program results. In fact, enforcing strict adherence to program protocols
in order to strengthen conclusions about program effects may actually
limit the ability to generalize those conclusions to less perfect, but more
typical program operations.
Ideally, randomized experiments in medicine are conducted as double-
blind studies, in which neither the subjects nor the researchers know who
is receiving the experimental treatment. However, double-blind studies in
social science are uncommon, making it hard sometimes to distinguish
the effects of a new program from the effects of introducing any novelty
into the classroom or workplace. Moreover, program staff may jeopardize
the random assignment process by exercising their own judgment in
recruiting and enrolling participants. Because of the critical importance of
the comparison groups’ equivalence for drawing conclusions about
program effects, it is important to check the effectiveness of random
assignment by comparing the groups’ equivalence on key characteristics
before program exposure.
Because of the difficulties in establishing a random process for assigning
units of study to a program, as well as the opportunity provided when only
a portion of the targeted population is exposed to the program, many
impact evaluations employ a quasi-experimental comparison group
design instead. This design also uses a treatment group and one or more
comparison groups; however, unlike the groups in the true experiment,
membership in these groups is not randomly assigned. Because the
groups were not formed through a random process, they may differ with
regard to other factors that affect their outcomes. Thus, it is usually not
possible to infer that the “raw” difference in outcomes between the groups
has been caused by the treatment. Instead, statistical adjustments such
as analysis of covariance should be applied to the raw difference to
compensate for any initial lack of equivalence between the groups.
Comparison groups may be formed from the pool of applicants who
exceed the number of program slots in a given locale or from similar
populations in other places, such as neighborhoods or cities, not served
by the program. Drawing on the research literature to identify the key
factors known to influence the desired outcomes will aid in forming
treatment and comparison groups that are as similar as possible, thus
strengthening the analyses’ conclusions. When the treatment group is
made up of volunteers, it is particularly important to address the potential
for “selection bias”—that is, that volunteers or those chosen to participate
Comparison Group Quasi-
experiments
Chapter 4: Designs for Assessing Program
Implementation and Effectiveness
Page 43 GAO-12-208G
will have greater motivation to succeed (for example, in attaining health,
education, or employment outcomes) than those who were not accepted
into the program. Statistical procedures, such as propensity score
analysis, are used to statistically model the variables that influence
participants’ assignment to the program and are then applied to analysis
of outcome data to reduce the influence of those variables on the
program’s estimated net impact. (For more information on propensity
scores, see Rosenbaum 2002.) However, in the absence of random
assignment, it is difficult to be sure that unmeasured factors did not
influence differences in outcomes between the treatment and comparison
groups.
A special type of comparison group design, regression discontinuity
analysis, compares outcomes for a treatment and control group that are
formed by having scores above or below a cut-point on a quantitative
selection variable rather than through random assignment. When
experimental groups are formed strictly on a cut-point and group
outcomes are analyzed for individuals close to the cut-point, the groups
can be left otherwise comparable except for the intervention. This
technique is often used where the persons considered most “deserving”
are assigned to the treatment, in order to address ethical concerns about
denying services to persons in need—for example, when additional
tutoring is provided only to children with the lowest reading scores. The
technique requires a quantitative assignment variable that users believe
is a credible selection criterion, careful control over assignment to ensure
that a strict cut-point is achieved, large sample sizes, and sophisticated
statistical analysis.
Both experiments and quasi-experiments can be difficult to implement
well in a variety of public settings. Confidence in conclusions about the
program’s impacts depends on ensuring that the treatment and
comparison groups’ experiences remain separate, intact, and distinct
throughout the life of the study so that any differences in outcomes can
be confidently attributed to the intervention. It is important to learn
whether control group participants access comparable treatment in the
community on their own. Their doing so could blur the distinction between
the two groups’ experiences. It is also preferred that treatment and control
group members not communicate, because knowing that they are being
treated differently might influence their perceptions of their experience
and, thus, their behavior.
Difficulties in Conducting
Comparison Group
Experiments
Chapter 4: Designs for Assessing Program
Implementation and Effectiveness
Page 44 GAO-12-208G
To resolve concerns about the ethics of withholding treatment widely
considered beneficial, members of the comparison group are usually
offered an alternative treatment or whatever constitutes common practice.
Thus, experiments are usually conducted to test the efficacy of new
programs or of new provisions or practices in an existing program. In this
case, however, the evaluation will no longer be testing whether a new
approach is effective at all; it will test whether it is more effective than
standard practice.
In addition, comparison group designs may not be practical for some
programs if the desired outcomes do not occur often enough to be
observed within a reasonable sample size or study length. Studies of
infrequent outcomes may require quite large samples to permit detection
of a difference between the experimental and control groups. Because of
the practical difficulties of maintaining intact experimental groups over
time, experiments are also best suited for assessing outcomes within 1 to
2 years after the intervention, depending on the circumstances.
Some federal programs and policies are not amenable to comparison
group designs because they are implemented all at once, all across the
country, with no one left untreated to serve in a comparison group. In
such instances, quasi-experimental single group designs compare the
outcomes for program participants before and after program exposure or
the outcomes associated with natural variation in program activities,
intensity or duration. In most instances, the simple version of a before-
and-after design does not allow causal attribution of observed changes to
exposure to the program because it is possible that other factors may
have influenced those outcomes during the same time.
Before-and-after designs can be strengthened by adding more
observations on outcomes. By taking many repeated observations of an
outcome before and after an intervention or policy is introduced, an
interrupted time-series analysis can be applied to the before-and-after
design to help draw causal inferences. Long data series are used to
smooth out the effects of random fluctuations over time. Statistical
modeling of simultaneous changes in important external factors helps
control for their influence on the outcome and, thus, helps isolate the
impact of the intervention. This approach is used for full-coverage
programs in which it may not be possible to find or form an untreated
comparison group. The need for lengthy data series means the technique
is used where the evaluator has access to long-term, detailed
government statistical series or institutional records. For example,
Statistical Analysis of
Observational Data
Chapter 4: Designs for Assessing Program
Implementation and Effectiveness
Page 45 GAO-12-208G
• To assess the effectiveness of a product safety regulation in reducing
injuries from a class of toys, the evaluator could analyze hospital
records of injuries associated with these toys for a few years both
before and after introduction of the regulation. To help rule out the
influence of alternative plausible explanations, the evaluator might
correlate these injury data with data on the size of the relevant age
group and sales of these toys over the same time period.
An alternative observational approach is a cross-sectional study that
measures the target population’s exposure to the intervention (rather than
controls its exposure) and compares the outcomes of individuals
receiving different levels of the intervention. Statistical analysis is used to
control for other plausible influences on the outcomes. Exposure to the
intervention can be measured by whether a person was enrolled or how
often a person participated in or was exposed to the program. This
approach is used with full-coverage programs for which it is impossible to
directly form treatment and control groups; nonuniform programs, in
which different individuals are exposed differently; and interventions in
which outcomes are observed too infrequently to make a prospective
study practical. For example,
• An individual’s annual risk of being in a car crash is so low that it
would be impractical to randomly assign (and monitor) thousands of
individuals to use (or not use) their seat belts in order to assess seat
belts’ effectiveness in preventing injuries during car crashes. Instead,
the evaluator can analyze data on seat belt use and injuries in car
crashes with other surveys on driver and passenger use of seat belts
to estimate the effectiveness of seat belts in reducing injury.
Although this paper describes process and outcome evaluations as if they
were mutually exclusive, in practice an evaluation may include multiple
design components to address separate questions addressing both
process and outcomes. In addition, comprehensive evaluations are often
designed to collect both process and outcome information in order to
understand the reasons for program performance and learn how to
improve results. For example,
• Evaluators analyze program implementation data to ensure that key
program activities are in place before collecting data on whether the
desired benefits of the activities have been achieved.
Comprehensive
Evaluations Explore Both
Process and Results
Chapter 4: Designs for Assessing Program
Implementation and Effectiveness
Page 46 GAO-12-208G
• Evaluations of program effectiveness also measure key program
components to help learn why a program is not working as well as
was expected.
An evaluation may find that a program failed to achieve its intended
outcomes for a variety of reasons, including: incomplete or poor quality
implementation of the program; problems in obtaining valid and reliable
data from the evaluation; environmental influences that blunt the
program’s effect; or the ineffectiveness of the program or intervention for
the population and setting in which it was tested. Thus, examination of
program implementation is very important to interpreting the results on
outcomes. Moreover, because an impact evaluation may be conducted in
a restricted range of settings in order to control for other influences on
outcomes, its findings may not apply to other settings or subgroups of
recipients. Thus, it is important to test the program or intervention’s
effects in several settings or under various circumstances before drawing
firm conclusions about its effectiveness. A formal synthesis of the findings
of multiple evaluations can provide important information about the
limitations on—or factors influencing—program impacts, and be
especially helpful in learning what works for whom and under what
circumstances.
As evaluation designs are tailored to the nature of the program and the
questions asked, it becomes apparent that certain designs are
necessarily excluded for certain types of programs. This is particularly
true of impact evaluations because of the stringent conditions placed on
the evidence needed to draw causal conclusions with confidence.
Experimental research designs are best adapted to assess discrete
interventions under carefully controlled conditions in the experimental
physical and social sciences. The federal government has only relatively
recently expanded its efforts to assess the effectiveness of all federal
programs and policies, many of which fail to meet the requirements for
successful use of experimental research designs.
To assist OMB officials in their efforts to assess agency evaluation efforts,
an informal network of federal agency evaluators provided guidance on
the relevance of various evaluation designs for different types of federal
programs. Table 5 summarizes the features of the designs discussed in
this chapter as well as the types of programs employing them.
Designs for Different
Types of Programs
Chapter 4: Designs for Assessing Program
Implementation and Effectiveness
Page 47 GAO-12-208G
Table 5: Designs for Assessing Effectiveness of Different Types of Programs
Typical design
Comparison controlling for
alternative explanations Best suited for
Process and outcome
monitoring or evaluation
Performance and preexisting goals or standards, such
as
• R&D criteria of relevance, quality, and performance
• productivity, cost effectiveness, and efficiency
standards
• customer expectations or industry benchmarks
Research, enforcement, information and
statistical programs, business-like enterprises,
and mature, ongoing programs where
• coverage is national and complete
• few, if any, alternatives explain observed
outcomes
Quasi-experiments: single
group
Outcomes for program participants before and after the
intervention:
• collects outcome data at multiple points in time
• statistical adjustments or modeling control for
alternative causal explanations
Regulatory and other programs where
• clearly defined interventions have distinct
starting times
• coverage is national and complete
• randomly assigning participants is NOT
feasible, practical, or ethical
Quasi-experiments:
comparison groups
Outcomes for program participants and a comparison
group closely matched to them on key characteristics:
• key characteristics are plausible alternative
explanations for a difference in outcomes
• measures outcomes before and after the
intervention (pretest, posttest)
Service and other programs where
• clearly defined interventions can be
standardized and controlled
• coverage is limited
• randomly assigning participants is NOT
feasible, practical, or ethical
Randomized experiments:
control groups
Outcomes for a randomly assigned treatment group and
a nonparticipating control group:
• measures outcomes preferably before and after the
intervention (pretest, posttest)
Service and other programs where
• clearly defined interventions can be
standardized and controlled
• coverage is limited
• randomly assigning participants is
feasible and ethical
Source Adapted from Bernholz et al. 2006.
Some types of federal programs, such as those funding basic research
projects or the development of statistical information, are not expected to
have readily measurable effects on their environment. Therefore,
research programs have been evaluated on the quality of their processes
and products and relevance to their customers’ needs, typically through
expert peer review of portfolios of completed research projects. For
example, the Department of Energy adopted criteria used or
recommended by OMB and the National Academy of Sciences to assess
research and development programs’ relevance, quality, and
performance (U.S. Department of Energy 2004.)
Regulatory and law enforcement programs can be evaluated according to
the level of compliance with the pertinent rule or achievement of desired
health or safety conditions, obtained through ongoing outcome
Chapter 4: Designs for Assessing Program
Implementation and Effectiveness
Page 48 GAO-12-208G
monitoring. The effectiveness of a new law or regulation might be
evaluated with a time-series design comparing health or safety conditions
before and after its enactment, while controlling for other possible
influences. Comparison group designs are not usually applied in this area
because of unwillingness to selectively enforce the law.
Experimental and quasi-experimental impact studies are better suited for
programs conducted on a small scale at selected locations, where
program conditions can be carefully controlled, rather than at the national
level. Such designs are particularly appropriate for demonstration
programs testing new approaches or initiatives, and are not well suited for
mature, universally available programs.
The next chapter outlines a number of approaches taken to evaluating
federal programs that are not well suited to these most common designs,
either because of the structure of the program or the context in which it
operates.
GAO. 1990. Case Study Evaluations, GAO/PEMD-10.1.9. Washington,
D.C. November.
GAO. 2003. Federal Programs: Ethnographic Studies Can Inform
Agencies’ Actions, GAO-03-455. Washington, D.C. March.
GAO. 2009. Program Evaluation: A Variety of Rigorous Methods Can
Help Identify Effective Interventions, GAO-10-30. Washington, D.C.
Nov. 23.
GAO. 2010. Streamlining Government: Opportunities Exist to Strengthen
OMB’s Approach to Improving Efficiency, GAO-10-394. Washington, D.C.
May 7.
Abt Associates and QED Group. 2004. Evaluation of the Welfare to Work
Voucher Program: Report to Congress. U.S. Department of Housing and
Urban Development, Office of Policy Development and Research. March.
For More Information
GAO documents
Other resources
Chapter 4: Designs for Assessing Program
Implementation and Effectiveness
Page 49 GAO-12-208G
Bernholz, Eric and others. 2006. Evaluation Dialogue Between OMB Staff
and Federal Evaluators: Digging a Bit Deeper into Evaluation Science.
Washington, D.C. July. http://www.fedeval.net/docs/omb2006briefing.pdf
Enders, Walter. 2009. Applied Econometric Time Series, 3rd ed.
Hoboken, N.J.: Wiley.
Langbein, Laura and Claire L. Felbinger. 2006. Public Program
Evaluation: A Statistical Guide. Armonk, N.Y.: M.E. Sharpe.
Lipsey, Mark W. “Theory as Method: Small Theories of Treatments.”
1993. New Directions for Program Evaluation 57:5-38. Reprinted in 2007,
New Directions for Evaluation 114:30-62.
OMB (U.S. Office of Management and Budget), Office of Information and
Regulatory Affairs. 2006. Standards and Guidelines for Statistical
Surveys. Washington, D.C. September.
http://www.whitehouse.gov/omb/inforeg_statpolicy#pr
Rosenbaum, Paul R. 2002. Observational Studies, 2nd ed. New York:
Springer.
Rossi, Peter H., Mark W. Lipsey, and Howard E. Freeman. 2004.
Evaluation: A Systematic Approach, 7th ed. Thousand Oaks, Calif.: Sage.
Shadish, William R., Thomas D. Cook, and Donald T. Campbell. 2002.
Experimental and Quasi-Experimental Designs for Generalized Causal
Inference. Boston: Houghton Mifflin.
Stake, Robert E. 1995. The Art of Case Study Research. Thousand Oaks,
Calif.: Sage.
U.S. Department of Energy. 2004. Peer Review Guide: Based on a
Survey of Best Practices for In-Progress Peer Review. Prepared by the
Office of Energy Efficiency and Renewable Energy Peer Review Task
Force. Washington, D.C. August.
http://www1.eere.energy.gov/ba/pba/pdfs/2004peerreviewguide.pdf.
Yin, Robert K. 2009. Case Study Research: Design and Methods, 4th ed.
Thousand Oaks, Calif.: Sage.
Chapter 5: Approaches to Selected
Methodological Challenges
Page 50 GAO-12-208G
Most of the impact designs discussed in chapter 4 were developed to test
hypotheses about the causal effects of individual factors or discrete
interventions on clearly defined outcomes. These designs may have
limited relevance and credibility on their own for assessing the effects of
federal programs where neither the intervention nor the desired outcome
is clearly defined or measured. In addition, many, if not most, federal
programs aim to improve some aspect of complex systems, such as the
economy or the environment, over which they have limited control, or
share responsibilities with other agencies for achieving their objectives.
Thus, it can be difficult to confidently attribute a causal connection
between the program and the observed outcomes. This chapter describes
some of the evaluation strategies that federal agencies have used to
develop performance information for these types of programs that can
inform management, oversight, and policy.
In many federal programs, it can be difficult to assess the program’s
effectiveness in achieving its ultimate objectives because it is difficult to
obtain data on those goals. This can occur because there is no common
measure of the desired outcome or because the desired benefits for the
public are not frequently observed.
A federal program might lack common national data on a desired
outcome because the program is relatively new, new to measuring
outcomes, or has limited control over how service providers collect and
store information. Where state programs operate without much federal
direction, outcome data are often not comparable across the states.
Federal agencies have taken different approaches to obtaining common
national outcome data, depending in part on whether such information is
needed on a recurring basis (GAO 2003):
• collaborating with others on a common reporting format;
• recoding state data into a common format;
• conducting a special survey to obtain nation-wide data.
Where federal programs operate through multiple local public or private
agencies, careful collaboration may be required to ensure that the data
they collect are sufficiently consistent to permit aggregation nationwide.
To improve the quality and availability of substance abuse prevention and
Chapter 5: Approaches to Selected
Methodological Challenges
Outcomes That Are
Difficult to Measure
Challenge: Lack of
Common Outcome
Measures
Collaborate with Others on a
Common Reporting Format
Chapter 5: Approaches to Selected
Methodological Challenges
Page 51 GAO-12-208G
treatment, the Substance Abuse and Mental Health Services
Administration (SAMHSA) awards block grants to states to help fund local
drug and alcohol abuse programs. In order to measure progress towards
national goals and the performance of programs administered by states’
substance abuse and mental health agencies, SAMHSA funded pilot
studies and collaborated with state agencies and service providers in
developing national outcome measures for an ongoing performance
monitoring system. The process of developing and agreeing upon data
definitions has taken several years, but allows them to assess
improvements in substance abuse treatment outcomes and monitor the
performance of SAMHSA block grants. SAMHSA has also invested in
states’ data infrastructure improvement activities such as software,
hardware, and training in how to use standardized data definitions (U.S.
Department of Health and Human Services n.d.).
Alternatively, if states already have their own distinct, mature data
systems, it may not be practical to expect those systems to adopt new,
common data definitions. Instead, to meet federal needs to assess
national progress, a federal agency may choose to support a special data
collection that abstracts data from state systems and recodes them into a
common format, permitting cross-state and national analyses. For
example, in order to analyze highway safety policies, the National
Highway Traffic Safety Administration has invested in a nationwide
system to extract data from state records to develop a well-accepted
national database on fatal automobile crashes. A standard codebook
provides detailed instructions on how to record data from state and local
emergency room and police records into a common format that can
support sophisticated analyses into the factors contributing to crashes
and associated fatalities (GAO 2003). Although such a data collection and
analysis system can be initially expensive to develop, it is likely to be less
expensive to maintain such a system, and much more practical than
attempting to gain agreements for data collection changes from hospitals
and police departments across the country.
Some federal agencies also, of course, conduct periodic sample surveys
or one-time studies to collect new data that supplements data from
existing performance reporting systems. For example, SAMHSA conducts
a voluntary periodic survey of specialty mental health organizations that
are not subject to the agency’s routine grantee reporting requirements
(U.S. Department of Health and Human Services n.d.). In addition, to
obtain information on drug abusers who are not in treatment, they
conduct an annual national household survey of drug use. Such surveys
Recode State Data into a
Common Format
Conduct a Special Survey to
Obtain Nation-Wide Data
Chapter 5: Approaches to Selected
Methodological Challenges
Page 52 GAO-12-208G
can provide valuable information about how well existing programs are
serving the population’s needs.
Some federal programs are created to respond to national concerns, such
as increased cancer rates or environmental degradation, which operate in
a lengthy time frame and are not expected to resolve quickly. Thus,
changes in intended long-term outcomes are unlikely to be observed
within an annual performance reporting cycle or even, perhaps, within a
five-year evaluation study. Other programs aim to prevent or provide
protection from events that are very infrequent and, most importantly, not
predictable, such as storms or terrorist attacks, for which it is impractical
to set annual or other relatively short-term goals. Evaluation approaches
to these types of programs may rely heavily on well-articulated program
logic models to depict the program’s activities as multi-step strategies for
achieving its goals. Depending on how infrequent or unexpected
opportunities may be to observe the desired outcome, an evaluator might
choose to:
• measure program effects on short-term or intermediate goals;
• assess the quality of an agency’s prevention or risk management
plan; or
• conduct a thorough after-action or critical-incident review of any
incidents that do occur.
To demonstrate progress towards the program’s ultimate goals, the
evaluator can measure the program’s effect on short-term and
intermediate outcomes that are considered important interim steps
towards achieving the program’s long-term goals. This approach is
particularly compelling when combined with findings from the research
literature that confirms the relationship of short-term goals (such as
increased vaccination rates) to the program’s long-term goals (such as
reduced incidence of communicable disease). (See GAO 2002 for
examples.) Moreover, tracking performance trends and progress towards
goals may provide timely feedback that can inform discussion of options
for responding to emerging performance problems.
Several federal programs are charged with managing risks that are
infrequent but potentially quite dangerous, in a wide array of settings:
Challenge: Desired
Outcomes Are Infrequently
Observed
Measure Effects on Short-Term
or Intermediate Goals
Assess the Quality of a
Prevention or Risk
Management Plan
Chapter 5: Approaches to Selected
Methodological Challenges
Page 53 GAO-12-208G
banking, intelligence, counter-terrorism, natural disasters, and community
health and safety. Generally, risk management involves:
• assessing potential threats, vulnerabilities of assets and networks,
and the potential economic or health and safety consequences;
• assessing and implementing countermeasures to prevent incidents
and reduce vulnerabilities to minimize negative consequences; and
• monitoring and evaluating their effectiveness (GAO 2005).
Depending on the nature of the threat, one federal program may focus
more on prevention (for example, of communicable disease) while
another focuses on response (for example, to hurricanes). Some threats
occur frequently enough that program effectiveness can be readily
measured as the reduction in threat incidents (such as car crashes) or
consequences (such as deaths and injuries). Where threat incidents do
not occur frequently enough to permit direct observation of the program’s
success in mitigating their consequences, evaluators have a couple
choices.
The evaluator could assess the effectiveness of a risk-management
program through assessing (1) how well the program followed the
recommended “best practices” of design, including conducting a
thorough, realistic assessment of threats and vulnerabilities, and cost-
benefit analysis of alternative risk reduction strategies; and (2) how
thoroughly the agency implemented its chosen strategy, such as installing
physical protections or ensuring staff are properly trained.
Alternatively, an evaluator may choose to conduct simulations or
exercises to assess how well an agency’s plans anticipate the nature of
its threats and vulnerabilities, as well as how well agency staff and
partners are prepared to carry out their responsibilities under their plans.
Exercises may be “table-top,” where officials located in an office respond
to virtual reports of an incident, or “live,” where volunteers act out the
roles of victims in public places to test the responses of emergency
services personnel. Exercises may be especially useful for obtaining a
realistic assessment of complex risk management programs that require
coordination among multiple agencies or public and private sector
organizations.
When a threat incident is observed, an evaluator can conduct an ‘after-
action’ or ‘critical incident’ review to assess the design and execution–or
Conduct an After-Action or
Critical-Incident Review
Chapter 5: Approaches to Selected
Methodological Challenges
Page 54 GAO-12-208G
effectiveness—of the prevention or risk mitigation program. The Army
developed after-action reviews as a training methodology for soldiers to
evaluate their performance against standards and develop insights into
their strengths, weaknesses, and training needs (U.S. Department of the
Army 1993). State and federal public safety agencies have adopted them
to identify ways to improve emergency response. These reviews consist
of a structured, open discussion of participants’ observations of what
occurred during an incident to develop ‘lessons learned’ about the
effectiveness of plans and procedures and actionable recommendations.
Reviews involve (1) detailed description of the nature and context of the
incident and the actions taken and resources used step-by-step; followed
by (2) a critique to assess whether plans and procedures were useful in
addressing the incident and provide suggestions for improvement. These
reviews may be formal—with an external facilitator or observer and a
written report to management—or informal—conducted as an internal
review to promote learning. Although identifying the factors contributing to
success or failure in handling an incident could provide useful insight into
the effectiveness of a risk mitigation program, the focus of these reviews
is primarily on learning rather than judging program effectiveness.
With increased interest in assuring accountability for the value of
government expenditures, have come increased efforts to demonstrate
and quantify the value of public investments in scientific research. An
evaluator might readily measure the effectiveness of an applied research
program by whether it met its goal to improve the quality, precision, or
efficiency of tools or processes. However, basic research programs do
not usually have such immediate, concrete goals. Instead, goals for
federal research programs can include advancing knowledge in a field,
and building capacity for future advances through developing useful tools
or supporting the scientific community. In addition, multiyear investments
in basic research might be expected to lead to innovations in technology
that will (eventually) yield social or financial value, such as energy
savings or security. (For more information about methods for assessing
these effects, see Ruegg and Jordan 2007.) Common agency
approaches to evaluating research programs include:
• external expert review of a research portfolio;
• bibliometric analyses of research citations and patents.
Challenge: Benefits of
Research Programs Are
Difficult to Predict
Chapter 5: Approaches to Selected
Methodological Challenges
Page 55 GAO-12-208G
To assess the quality of their research programs and obtain program
planning advice, the National Science Foundation (NSF) adopted an
external expert review process called a Committee of Visitors (COV)
review. Periodically, panels of independent experts review the technical
and managerial stewardship of a specific program (a portfolio of research
projects), compare plans with progress made, and evaluate the outcomes
to assess their contributions to NSF’s mission and goals. COV reviews
provide external expert judgments on 1) assessments of the quality and
integrity of program operations and program-level technical and
managerial matters pertaining to project decisions; and 2) comments on
how the outputs and outcomes generated by awardees have contributed
to NSF’s mission and strategic outcome goals. Other federal science
agencies have adopted similar expert panel reviews as independent
evaluations of their basic research programs (U.S. Department of Energy
2004).
Since publications and patents constitute major outputs of research
programs and large databases capture these outputs, bibliometric
analysis of research citations or patents is a popular way of assessing the
productivity of research. In addition to simply tracking the quantity of
publications, analysis of where, how often and by whom the papers are
cited can provide information about the perceived relevance, impact and
quality of the papers and can identify pathways of information flow.
Many federal programs are not discrete interventions aiming to achieve a
specific outcome but, instead, efforts to improve complex systems over
which they have limited control. Moreover, in the United States, federal
and state governments often share responsibility for the direction of
federal programs, so a federal program may not represent a uniform
package of activities or services across the country.
Federal grant programs vary greatly as to whether they have performance
objectives or a common set of activities across grantees such as state
and local agencies or nonprofit service providers. Where a grant program
represents a discrete program with a narrow set of activities and
performance-related objectives, such as a food delivery program for
seniors, it can often be evaluated with the methods described in chapter
4. However, a formula or ‘block’ grant, with loosely defined objectives that
simply adds to a stream of funds supporting ongoing state or local
programs, presents a significant challenge to efforts to portray the results
External Expert Portfolio
Review
Bibliometric Analysis
Complex Federal
Programs and
Initiatives
Challenge: Benefits of
Flexible Grant Programs
Are Difficult to Summarize
Chapter 5: Approaches to Selected
Methodological Challenges
Page 56 GAO-12-208G
of the federal or ‘national’ program (GAO 1998a). Agencies have
deployed a few distinct approaches, often in combination:
• describe national variation in local approaches;
• measure national improvement in common outputs or outcomes;
• conduct effectiveness evaluations in a sample of sites.
An important first step in evaluating the performance of flexible grant
programs is to describe the variation in approaches deployed locally,
characteristics of the population served, and any information available on
service outputs or outcomes. Depending on the nature of grantee
reporting requirements, this information might be obtained from a review
of federal program records or require a survey of grantees or local
providers. Such descriptive information can be valuable in assessing how
well the program met Congress’ intent for the use and beneficiaries of
those funds. In addition, where there is prior research evidence on the
effectiveness of particular practices, this descriptive data can provide
information, at least, on the extent to which grantees are deploying
effective or ‘research-based’ practices.
Where the federal grant program has performance-related objectives but
serves as a funding stream to support and improve the capacity of a state
function or service delivery system, state (but not uniquely federal)
program outcomes can be evaluated by measuring aggregate
improvements in the quality of or access to services, outreach to the
targeted population, or participant outcomes over time. Depending on the
program, this information may be collected as part of state program
administration, or require special data collection to obtain comparable
data across states. For example, the Department of Education’s National
Assessment of Educational Progress tests a cross-sectional sample of
children on a variety of key subjects, including reading and math, and
regularly publishes state-by-state data on a set of common outcome
measures. These national data also provide a comparative benchmark for
the results of states’ own assessments (Ginsburg and Rhett 2003).
However, because cross-sectional surveys lack information linking
specific use of federal funds to expected outcomes, they cannot assess
the effectiveness of federal assistance in contributing to those service
improvements; identifying those links is often very difficult in grant
programs of this type.
Describe National Variation in
Local Approaches
Measure National Improvement
in Common Outputs or
Outcomes
Chapter 5: Approaches to Selected
Methodological Challenges
Page 57 GAO-12-208G
Some federal grant programs support distinct local projects to stimulate or
test different approaches for achieving a performance objective. To
assess such programs, the evaluator might study a sample of projects to
assess their implementation and effectiveness in meeting their objectives.
Individual impact evaluations might be arranged for as part of the original
project grants, or conducted as part of a nationally-directed evaluation.
Sites for evaluation might be selected purposively, to test the
effectiveness of a variety of promising program approaches or represent
the range in quality of services nationally (Herrell and Straw 2002).
For example, cluster evaluations, as used by the W. K. Kellogg
Foundation, examine a loosely connected set of studies of community-
based initiatives to identify common themes or components associated
with positive impacts, and the reasons for such associations (W. K.
Kellogg Foundation 2004). Cluster evaluations examine evidence of
individual project effectiveness but do not aggregate that data across
studies. Multisite evaluations, as frequently seen in federally-funded
programs, may involve variation across sites in interventions and
measures of project effectiveness, but typically use a set of common
measures to estimate the effectiveness of the interventions and examine
variation across sites in outcomes. (See discussion of comprehensive
evaluations in chapter 4.) Both of these evaluation approaches are quite
different from a multicenter clinical trial (or impact study) that conducts
virtually the same intervention and evaluation in several sites to test the
robustness of the approach’s effects across sites and populations (Herrell
and Straw 2002).
Case study evaluations, through providing more in-depth information
about how a federal program operates in different circumstances, can
serve as valuable supplements to broad surveys when specifically
designed to do so. Case studies can be designed to follow-up on low or
high performers, in order to explain–or generate hypotheses about—what
is going on and why.
In contrast to programs that support a particular set of activities aimed at
achieving a specified objective, some comprehensive reform initiatives
may call for collective, coordinated actions in communities in multiple
areas such as altering public policy, improving service practice, or
engaging the public to create system reform. This poses challenges to the
evaluator in identifying the nature of the intervention (or program), the
desired outcomes, as well as an estimate of what would have occurred in
the absence of these reforms. Depending on the extent to which the
Conduct Effectiveness
Evaluations in a Sample of
Sites
Challenge: Assess the
Progress and Results of
Comprehensive Reforms
Chapter 5: Approaches to Selected
Methodological Challenges
Page 58 GAO-12-208G
dimensions of reform are well understood, the progress of reforms might
be measured quantitatively in a survey or through a more exploratory
form of case study.
For example, in the Department of Education’s Comprehensive School
Reform demonstration program, federal grantees were encouraged to
strengthen several aspects of school operations–-such as curriculum,
instruction, teacher development, parental involvement—and to select
and adopt models that had been found effective in other schools, in an
effort to improve student achievement. The comprehensive evaluation of
this program used three distinct methodological approaches to answer
distinct questions about implementation and effects (U.S. Department of
Education 2010)
1. Multivariate statistical analyses comparing grantees with matched
comparison schools to determine whether receiving a grant was
associated with student achievement level increases three to five
years later;
2. Quantitative descriptive analyses of reform implementation from a
survey of principals and teachers in a random sample of grantees and
matched comparison schools to determine the comprehensiveness of
reform implementation; and
3. Qualitative case study analyses to study reform component
implementation and understand the process by which chronically low-
performing schools turned themselves around and sustained student
achievement gains.
Note that because a school reform effort by design applies to everyone in
the school, the evaluators formed a comparison group by matching each
grantee school with a school in another community with similar socio-
economic characteristics. Moreover, this study’s analyses of the schools’
reforms were greatly assisted by being able to draw on the set of potential
reforms listed in the legislation.
A different approach is required for a much more open-ended program,
such as the Department of Housing and Urban Development’s
Empowerment Zones and Enterprise Communities Program. This
program provided grants and tax incentives to economically
disadvantaged communities which were encouraged to develop their own
individual economic development strategies around four key principles:
economic opportunity, sustainable community development, community-
Follow-up Survey Findings
with Case Studies
Conduct Exploratory Case
Studies
Chapter 5: Approaches to Selected
Methodological Challenges
Page 59 GAO-12-208G
based partnerships, and a strategic vision for change. Local evaluators
assisted in collecting data in each of 18 case study sites to track how
each community organized itself, set goals, and developed and
implemented plans to achieve those goals–its theory of change
(Fulbright-Anderson et al. 1998).
Case studies are recommended for assessing the effectiveness of
comprehensive reforms that are so deeply integrated with the context
(i.e., community) that no truly adequate comparison case can be found.
In-depth interviews and observations are used to capture the changes in
and relationships between processes, while outcomes may be measured
quantitatively. The case study method is used to integrate this data into a
coherent picture or story of what was achieved and how. In programs that
are more direct about what local reform efforts are expected to achieve,
the evaluator might provide more credible support for conclusions about
program effects by: (1) making specific, refutable predictions of program
effects, and (2) introducing controls for, or providing strong arguments
against, other plausible explanations for observed outcomes. This theory
of change approach cannot provide statistical estimates of effect sizes,
but can provide detailed descriptions of the unfolding of the intervention
and potential explanations for how and why the process worked to
produce outcomes (Fulbright-Anderson et al. 1998, Yin and Davis 2007).
Attributing observed changes in desired outcomes to the effect of a
program requires ruling out other plausible explanations for those
changes. Environmental factors such as historical trends in community
attitudes towards smoking could explain changes in youths’ smoking
rates over time. Other programs funded with private, state, or other
federal funds may also strive for similar goals to the program being
evaluated. Although random assignment of individuals to treatment and
comparison groups is intended to cancel out the influence of those
factors, in practice, the presence of these other factors may still blur the
effect of the program of interest or randomization may simply not be
feasible. Collecting additional data and targeting comparisons to help rule
out alternative explanations can help strengthen conclusions about an
intervention’s impact from both randomized and nonrandomized designs
(GAO 2009, Mark and Reichardt 2004).
In general, to help isolate the impact of programs aimed at the same goal
it can be useful to construct a logic model for each program—carefully
specifying the programs’ distinct target audiences and expected short-
term outcomes—and to assess the extent to which the programs actually
Challenge: Isolating Impact
When Several Programs
Are Aimed at the Same
Outcome
Chapter 5: Approaches to Selected
Methodological Challenges
Page 60 GAO-12-208G
operate in the same localities and reach the same populations. Then the
evaluator can devise a data collection approach or set of comparisons
that could isolate the effects of the distinct programs, such as
• narrow the scope of the outcome measure;
• measure additional outcomes not expected to change;
• test hypothesized relationships between the programs.
Some programs have strategic goals that imply that they have a more
extensive or broader range than they in fact do. By clarifying very
specifically the program’s target audience and expected behavior
changes, the evaluator can select an outcome measure that is closely
tailored to the most likely expected effects of the program and distinguish
those effects from those of other related programs.
For example, to distinguish one antidrug media campaign from other
antidrug messages in the environment, the campaign used a distinctive
message to create a brand that would provide a recognizable element
and improve recall. Then, the evaluation’s survey asked questions about
recognition of the brand, attitudes, and drug use so that analysis could
correlate attitudes and behavior changes with exposure to this particular
campaign (GAO 2002, Westat 2003).
In another example, the large number of workplaces in the country makes
it impractical for the Occupational Safety and Health Administration to
routinely perform health and safety inspections in all workplaces. Instead,
program officials indicated that they target their activities to where they
see the greatest problems—industries and occupations with the highest
rates of fatality, injury, or illness. Thus, the agency set a series of
performance goals that reflect differences in their expected influence,
setting goals for reductions in three of the most prevalent injuries and
illnesses and for injuries and illness in five “high-hazard” industries (GAO
1998b).
Another way to attempt to rule out plausible alternative explanations for
observed results is to measure additional outcomes that a treatment or
intervention is not expected to influence but arguably would be influenced
under alternative explanations for the observed outcomes. If one can
predict a relatively unique pattern of outcomes for the intervention, in
contrast to the alternative, and if the study confirms that pattern, then the
Narrow the Scope of the
Outcome Measure
Measure Additional Outcomes
Not Expected to Change
Chapter 5: Approaches to Selected
Methodological Challenges
Page 61 GAO-12-208G
alternative explanation becomes less plausible. In a simple example, one
can extend data collection either before or after the intervention to help
rule out the influence of unrelated historical trends on the outcome of
interest. If the outcome measure began to change before the intervention
could have plausibly have affected it, then that change was probably
influenced by some other factor.
Some programs aimed at similar broad outcomes may be expected also
to affect other programs. For example, the effectiveness of one program
that aims to increase the number of medical personnel in locations
considered medically underserved might be critical to ensuring that a
second program to increase the number of patients with health insurance
will result in their patients obtaining greater access to care. To assess the
effectiveness of the health insurance program, the evaluator could survey
potential recipients in a variety of locations where some are considered
medically underserved and some are not. Interviews could follow-up on
these hypotheses by probing reasons why potential recipients may have
had difficulty obtaining needed health care.
GAO. 1998a. Grant Programs: Design Features Shape Flexibility,
Accountability, and Performance Information, GAO/GGD-98-137.
Washington, D.C. June 22.
GAO. 1998b. Managing for Results: Measuring Program Results That Are
Under Limited Federal Control, GAO/GGD-99-16. Washington, D.C.
Dec. 11.
GAO. 2003. Program Evaluation: An Evaluation Culture and Collaborative
Partnerships Help Build Agency Capacity, GAO-03-454. Washington,
D.C. May 2.
GAO. 2009. Program Evaluation: A Variety of Rigorous Methods Can
Help Identify Effective Interventions, GAO-10-30. Washington, D.C.
Nov. 23.
GAO. 2002. Program Evaluation: Strategies for Assessing How
Information Dissemination Contributes to Agency Goals, GAO-02-923.
Washington, D.C. Sept. 30.
Test Hypothesized
Relationships between
Programs
For More Information
GAO documents
Chapter 5: Approaches to Selected
Methodological Challenges
Page 62 GAO-12-208G
GAO. 2005. Risk Management: Further Refinements Needed to Assess
Risks and Prioritize Protective Measures at Ports and Other Critical
Infrastructure. GAO-06-91. Washington, D.C. Dec. 15.
Domestic Working Group, Grant Accountability Project. 2005. Guide to
Opportunities for Improving Grant Accountability. Washington, D.C.: U.S.
Environmental Protection Agency, Office of Inspector General, October.
www.epa.gov/oig/dwg/index.htm.
Fulbright-Anderson, Karen, Anne C. Kubisch, and James P. Connell, eds.
1998. New Approaches to Evaluating Community Initiatives. vol. 2.
Theory, Measurement, and Analysis. Washington, D.C.: The Aspen
Institute.
Ginsburg, Alan, and Nancy Rhett. 2003. “Building a Better Body of
Evidence: New Opportunities to Strengthen Evaluation Utilization.”
American Journal of Evaluation 24: 489–98.
Herrell, James M., and Roger B. Straw, eds. 2002. Conducting Multiple
Site Evaluations in Real-World Settings. New Directions for Evaluation
94. San Francisco: Jossey-Bass, Summer.
Mark, Melvin M. and Charles S. Reichardt. 2004. “Quasi-Experimental
and Correlational Designs: Methods for the Real World When Random
Assignment Isn’t Feasible.” In Carol Sansone, Carolyn C. Morf, and A. T.
Panter, eds. The Sage Handbook of Methods in Social Psychology.
Thousand Oaks, Calif.: Sage.
Ruegg, Rosalie, and Gretchen Jordan. 2007. Overview of Evaluation
Methods for R&D Programs: A Directory of Evaluation Methods Relevant
to Technology Development Programs. Prepared under contract DE-
AC0494AL8500. Washington, D.C.: U.S. Department of Energy, Office of
Energy Efficiency and Renewable Energy. March.
U.S. Department of Education, Office of Planning, Evaluation and Policy
Development, Policy and Program Studies Service. 2010. Evaluation of
the Comprehensive School Reform Program Implementation and
Outcomes: Fifth Year Report. Washington, D.C.
U.S. Department of Energy. 2004. Peer Review Guide: Based on a
Survey of Best Practices for In-Progress Peer Review. Prepared by the
Office of Energy Efficiency and Renewable Energy Peer Review Task
Other resources
Chapter 5: Approaches to Selected
Methodological Challenges
Page 63 GAO-12-208G
Force. Washington, D.C.: August.
http://www1.eere.energy.gov/ba/pba/pdfs/2004peerreviewguide.pdf.
U.S. Department of Health and Human Services, Substance Abuse and
Mental Health Services Administration. n.d. SAMHSA Data Strategy: FY
2007- FY2011. Washington, D.C.
U.S. Department of Homeland Security, Federal Emergency Management
Agency, U.S. Fire Administration. 2008. Special Report: The After-Action
Critique: Training Through Lessons Learned. Technical Report Series.
USFA-TR-159. Emmitsburg, Md.: April.
U.S. Department of the Army, Headquarters. 1993. A Leader’s Guide to
After-Action Reviews, Training Circular 25-20. Washington, D.C.:
September 30. http://www.au.af.mil/au/awc/awcgate
W. K. Kellogg Foundation. 2004. W. K. Kellogg Foundation Evaluation
Handbook. Battle Creek, Mich.: Jan. 1, 1998, updated.
http://www.wkkf.org/knowledge-center/resources/2010/W-K-Kellogg-
Foundation-Evaluation-Handbook.aspx
Westat. 2003. Evaluation of the National Youth Anti-Drug Media
Campaign: 2003 Report of Findings. Prepared under contract N01DA-8-
5063. Rockville, Md.: National Institutes of Health, National Institute on
Drug Abuse, Dec. 22.
Yin, Robert K. and Darnella Davis 2007. “Adding New Dimensions to
Case Study Evaluations: The Case of Evaluating Comprehensive
Reforms.” New Directions for Evaluation 113:75-93.
Appendix I: Evaluation Standards
Page 64 GAO-12-208G
Different auditing and evaluation organizations have developed guidelines
or standards to help ensure the quality, credibility, and usefulness of
evaluations. Some standards pertain specifically to the evaluator’s
organization (for example, auditor independence), the planning process
(for example, stakeholder consultations), or reporting (for example,
documenting assumptions and procedures). While the underlying
principles substantially overlap, the evaluator will need to determine the
relevance of each guideline to the evaluator’s organizational affiliation
and the specific evaluation’s scope and purpose.
GAO publishes generally accepted government auditing standards
(GAGAS) for the use of individuals in government audit organizations
conducting a broad array of work, including financial and performance
audits. The standards are broad statements of auditors’ (or evaluators’)
responsibilities in an overall framework for ensuring that they have the
competence, integrity, objectivity, and independence needed to plan,
conduct, and report on their work. The standards use “performance audit”
to refer to “an independent assessment of the performance and
management of government programs against objective criteria or an
assessment of best practices and other information”; thus, it is intended to
include program process and outcome evaluations.
The general standards applying to all financial and performance audits
include the independence of the audit organization and its individual
auditors; the exercise of professional judgment; competence of staff; and
the presence of quality control systems and external peer reviews. The
field work standards for performance audits relate to planning the audit;
supervising staff; obtaining sufficient, competent, and relevant evidence;
and preparing audit documentation.
GAO. 2011. Government Auditing Standards: 2011 Internet Version.
Washington, D.C.: August. http://www.gao.gov/govaud/iv2011gagas.pdf
GAO’s transfer paper The Evaluation Synthesis lists illustrative questions
for assessing the soundness of each study’s basic research design,
conduct, analysis, and reporting—regardless of the design employed. The
questions address the clarity and appropriateness of study design,
measures, and analyses and the quality of the study’s execution and
reporting.
Appendix I: Evaluation Standards
“Yellow Book” of
Government Auditing
Standards
GAO’s Evaluation
Synthesis
Appendix I: Evaluation Standards
Page 65 GAO-12-208G
GAO.1992. The Evaluation Synthesis, revised, GAO/PEMD-10.1.2.
Washington, D.C.: March.
The American Evaluation Association (AEA) is a professional association
with U.S. headquarters for evaluators of programs, products, personnel,
and policies. AEA developed guiding principles for the work of
professionals in everyday practice and to inform evaluation clients and
the general public of expectations for ethical behavior. The principles are
broad statements of evaluators’ responsibilities in five areas: systematic
inquiry; competence; honesty and integrity; respect for people; and
responsibilities for general and public welfare.
AEA. 2004. Guiding Principles for Evaluators. July.
http://www.eval.org/Publications/GuidingPrinciples.asp.
A consortium of professional organizations (including the American
Evaluation Association), the Joint Committee on Standards for
Educational Evaluation, developed a set of standards for evaluations of
educational programs, which have been approved as an American
National Standard. The standards are organized into five major areas of
concern: to ensure program stakeholders find evaluations valuable
(utility); to increase evaluation effectiveness and efficiency (feasibility); to
support what is proper, fair, legal, right, and just in evaluations (propriety);
to increase the dependability and truthfulness of evaluation
representations and findings (accuracy); and to encourage accurate
documentation and a focus on improvement and accountability of
evaluation processes and products (evaluation accountability).
Yarbrough, D. B., L. M. Shulha, R. K. Hopson, and F. A. Caruthers. 2011.
The Program Evaluation Standards: A Guide for Evaluators and
Evaluation Users, 3rd ed. Thousand Oaks, Calif.: Sage.
American Evaluation
Association Guiding
Principles for
Evaluators
Program Evaluation
Standards, Joint
Committee on
Standards for
Educational
Evaluation
Appendix II: GAO Contact and Staff
Acknowledgments
Page 66 GAO-12-208G
Nancy Kingsbury (202) 512-2700 or kin[email protected]
In addition to the person named above, Stephanie Shipman, Assistant
Director, made significant contributions to this report. Additional
contributors include Thomas Clarke, Timothy Guinane, Penny Pickett,
and Elaine Vaurio.
Appendix II: GAO Contact and Staff
Acknowledgments
GAO Contact
Staff
Acknowledgments
Other Papers in This Series
Page 67 GAO-12-208G
Assessing the Reliability of Computer-Processed Data, external
version 1, GAO-09-680G. Washington, D.C.: July 2009.
Case Study Evaluations, GAO/PEMD-10.1.9, November 1990.
How to Get Action on Audit Recommendations, OP-9.2.1, July 1991.
Performance Measurement and Evaluation: Definitions and
Relationships, GAO-11-646SP, May 2011.
Prospective Evaluation Methods: The Prospective Evaluation Synthesis,
GAO/PEMD-10.1.10, November 1990.
Quantitative Data Analysis: An Introduction, GAO/PEMD-10.1.11,
May 1992.
Record Linkage and Privacy: Issues in Creating New Federal Research
and Statistical Information, GAO-01-126SP, April 2001.
The Evaluation Synthesis, revised, GAO/PEMD-10.1.2, March 1992.
The Results Act: An Evaluator’s Guide to Assessing Agency Annual
Performance Plans, version 1, GAO/GGD-10.1.20, April 1998.
Using Statistical Sampling, revised, GAO/PEMD-10.1.6, May 1992.
Using Structured Interviewing Techniques, GAO/PEMD-10.1.5,
June 1991.
Other Papers in This Series
(460621)
The Government Accountability Office, the audit, evaluation, and
investigative arm of Congress, exists to support Congress in meeting its
constitutional responsibilities and to help improve the performance and
accountability of the federal government for the American people. GAO
examines the use of public funds; evaluates federal programs and
policies; and provides analyses, recommendations, and other assistance
to help Congress make informed oversight, policy, and funding decisions.
GAO’s commitment to good government is reflected in its core values of
accountability, integrity, and reliability.
The fastest and easiest way to obtain copies of GAO documents at no
cost is through GAO’s website (www.gao.gov). Each weekday afternoon,
GAO posts on its website newly released reports, testimony, and
correspondence. To have GAO e-mail you a list of newly posted products,
go to www.gao.gov and select “E-mail Updates.”
The price of each GAO publication reflects GAO’s actual cost of
production and distribution and depends on the number of pages in the
publication and whether the publication is printed in color or black and
white. Pricing and ordering information is posted on GAO’s website,
http://www.gao.gov/ordering.htm.
Place orders by calling (202) 512-6000, toll free (866) 801-7077, or
TDD (202) 512-2537.
Orders may be paid for using American Express, Discover Card,
MasterCard, Visa, check, or money order. Call for additional information.
Connect with GAO on Facebook, Flickr, Twitter, and YouTube.
Subscribe to our RSS Feeds or E-mail Updates. Listen to our Podcasts .
Visit GAO on the web at www.gao.gov.
Contact:
Website: www.gao.gov/fraudnet/fraudnet.htm
E-mail: f[email protected]
Automated answering system: (800) 424-5454 or (202) 512-7470
Katherine Siggerud, Managing Director, [email protected], (202) 512-
4400, U.S. Government Accountability Office, 441 G Street NW, Room
7125, Washington, DC 20548
Chuck Young, Managing Director, y[email protected], (202) 512-4800
U.S. Government Accountability Office, 441 G Street NW, Room 7149
Washington, DC 20548
GAO’s Mission
Obtaining Copies of
GAO Reports and
Testimony
Order by Phone
Connect with GAO
To Report Fraud,
Waste, and Abuse in
Federal Programs
Congressional
Relations
Public Affairs
Please Print on Recycled Paper.
