This is the accessible text file for GAO report number GAO-10-30
entitled 'Program Evaluation: A Variety of Rigorous Methods Can Help
Identify Effective Interventions' which was released on November 23,
2009.
This text file was formatted by the U.S. Government Accountability
Office (GAO) to be accessible to users with visual impairments, as part
of a longer term project to improve GAO products' accessibility. Every
attempt has been made to maintain the structural and data integrity of
the original printed product. Accessibility features, such as text
descriptions of tables, consecutively numbered footnotes placed at the
end of the file, and the text of agency comment letters, are provided
but may not exactly duplicate the presentation or format of the printed
version. The portable document format (PDF) file is an exact electronic
replica of the printed version. We welcome your feedback. Please E-mail
your comments regarding the contents or accessibility features of this
document to Webmaster@gao.gov.
This is a work of the U.S. government and is not subject to copyright
protection in the United States. It may be reproduced and distributed
in its entirety without further permission from GAO. 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.
Report to Congressional Requesters:
United States Government Accountability Office:
GAO:
November 2009:
Program Evaluation:
A Variety of Rigorous Methods Can Help Identify Effective
Interventions:
GAO-10-30:
GAO Highlights:
Highlights of GAO-10-30, a report to congressional requesters.
Why GAO Did This Study:
Recent congressional initiatives seek to focus funds for certain
federal social programs on interventions for which randomized
experiments show sizable, sustained benefits to participants or
society. The private, nonprofit Coalition for Evidence-Based Policy
undertook the Top Tier Evidence initiative to help federal programs
identify interventions that meet this standard.
GAO was asked to examine (1) the validity and transparency of the
Coalition’s process, (2) how its process compared to that of six
federally supported efforts to identify effective interventions, (3)
the types of interventions best suited for assessment with randomized
experiments, and (4) alternative rigorous methods used to assess
effectiveness. GAO reviewed documents, observed the Coalition’s
advisory panel deliberate on interventions meeting its top tier
standard, and reviewed other documents describing the processes the
federally supported efforts had used. GAO reviewed the literature on
evaluation methods and consulted experts on the use of randomized
experiments.
The Coalition generally agreed with the findings. The Departments of
Education and Health and Human Services provided technical comments on
a draft of this report. The Department of Justice provided no comments.
What GAO Found:
The Coalition’s Top Tier Evidence initiative criteria for assessing
evaluation quality conform to general social science research
standards, but other features of its overall process differ from common
practice for drawing conclusions about intervention effectiveness. The
Top Tier initiative clearly describes how it identifies candidate
interventions but is not as transparent about how it determines whether
an intervention meets the top tier criteria. In the absence of detailed
guidance, the panel defined sizable and sustained effects through case
discussion. Over time, it increasingly obtained agreement on whether an
intervention met the top tier criteria.
The major difference in rating study quality between the Top Tier and
the six other initiatives examined is a product of the Top Tier
standard as set out in certain legislative provisions: the other
efforts accept well-designed, well-conducted, nonrandomized studies as
credible evidence. The Top Tier initiative’s choice of broad topics
(such as early childhood interventions), emphasis on long-term effects,
and use of narrow evidence criteria combine to provide limited
information on what is effective in achieving specific outcomes. The
panel recommended only 6 of 63 interventions reviewed as providing “
sizeable, sustained effects on important outcomes.” The other
initiatives acknowledge a continuum of evidence credibility by
reporting an intervention’s effectiveness on a scale of high to low
confidence.
The program evaluation literature generally agrees that well-conducted
randomized experiments are best suited for assessing effectiveness when
multiple causal influences create uncertainty about what caused
results. However, they are often difficult, and sometimes impossible,
to carry out. An evaluation must be able to control exposure to the
intervention and ensure that treatment and control groups’ experiences
remain separate and distinct throughout the study.
Several rigorous alternatives to randomized experiments are considered
appropriate for other situations: quasi-experimental comparison group
studies, statistical analyses of observational data, and—in some
circumstances—in-depth case studies. The credibility of their estimates
of program effects relies on how well the studies’ designs rule out
competing causal explanations. Collecting additional data and targeting
comparisons can help rule out other explanations.
GAO concludes that:
* requiring evidence from randomized studies as sole proof of
effectiveness will likely exclude many potentially effective and
worthwhile practices;
* reliable assessments of evaluation results require research expertise
but can be improved with detailed protocols and training;
* deciding to adopt an intervention involves other considerations in
addition to effectiveness, such as cost and suitability to the local
community; and;
* improved evaluation quality would also help identify effective
interventions.
What GAO Recommends:
GAO makes no recommendations.
View [hyperlink, http://www.gao.gov/products/GAO-10-30] or key
components. For more information, contact Nancy Kingsbury at (202) 512-
2700 or kingsburyn@gao.gov.
[End of section]
Contents:
Letter:
Background:
Top Tier Initiative's Process Is Mostly Transparent:
Top Tier Follows Rigorous Standards but Is Limited for Identifying
Effective Interventions:
Randomized Experiments Can Provide the Most Credible Evidence of
Effectiveness under Certain Conditions:
Rigorous Alternatives to Random Assignment Are Available:
Concluding Observations:
Agency and Third-Party Comments:
Appendix I: Steps Seven Evidence-Based Initiatives Take to Identify
Effective Interventions:
Appendix II: Comments from the Coalition for Evidence-Based Policy:
Appendix III: GAO Contact and Staff Acknowledgments:
Bibliography:
Related GAO Products:
Abbreviations:
AHRQ: Agency for Healthcare Research and Quality:
CDC: Centers for Disease Control and Prevention:
EPC: Evidence-based Practice Centers:
GPRA: Government Performance and Results Act of 1993:
HHS: Department of Health and Human Services:
MPG: Model Programs Guide:
NREPP: National Registry of Evidence-based Programs and Practices:
OMB: Office of Management and Budget:
PART: Program Assessment Rating Tool:
PRS: HIV/AIDS Prevention Research Synthesis:
SAMHSA: Substance Abuse and Mental Health Administration:
SCHIP: State Children's Health Insurance Program:
WWC: What Works Clearinghouse:
[End of section]
United States Government Accountability Office:
Washington, DC 20548:
November 23, 2009:
The Honorable Joseph I. Lieberman:
Chairman:
The Honorable Susan M. Collins:
Ranking Member:
Committee on Homeland Security and Governmental Affairs:
United States Senate:
The Honorable Mary L. Landrieu:
Chairman:
Subcommittee on Disaster Recovery:
Committee on Homeland Security and Governmental Affairs:
United States Senate:
Several recent congressional initiatives seek to focus funds in certain
federal social programs on activities for which the evidence of
effectiveness is rigorous--specifically, well-designed randomized
controlled trials showing sizable, sustained benefits to program
participants or society. To help agencies, grantees, and others
implement the relevant legislative provisions effectively, the private,
nonprofit Coalition for Evidence-Based Policy launched the Top Tier
Evidence initiative in 2008 to identify and validate social
interventions meeting the standard of evidence set out in these
provisions. In requesting this report, you expressed interest in
knowing whether limiting the search for effective interventions to
those that had been tested against these particular criteria might
exclude from consideration other important interventions. To learn
whether the Coalition's approach could be valuable in helping federal
agencies implement such funding requirements, you asked GAO to
independently assess the Coalition's approach. GAO's review focused on
the following questions.
1. How valid and transparent is the process the Coalition used--
searching, selecting, reviewing, and synthesizing procedures and
criteria--to identify social interventions that meet the standard of
"well-designed randomized controlled trials showing sizable, sustained
effects on important outcomes"?
2. How do the Coalition's choices of procedures and criteria compare to
(a) generally accepted design and analysis techniques for identifying
effective interventions and (b) similar standards and processes other
federal agencies use to evaluate similar efforts?
3. What types of interventions do randomized controlled experiments
appear to be best suited to assessing effectiveness?
4. For intervention types for which randomized controlled experiments
appear not to be well suited, what alternative forms of evaluation are
used to successfully assess effectiveness?
To assess the Coalition's Top Tier initiative, we reviewed documents,
conducted interviews, and observed the deliberations of its advisory
panel, who determined which interventions met the "top tier" evidence
standard--well-designed, randomized controlled trials showing sizable,
sustained benefits to program participants or society. We evaluated the
transparency of the initiative's process against its own publicly
stated procedures and criteria, including the top tier evidence
standard. To assess the validity of the Coalition's approach, we
compared its procedures and criteria to those recommended in program
evaluation textbooks and related publications, as well as to the
processes actually used by six federally supported initiatives with a
similar purpose to the Coalition. Through interviews and database
searches, we identified six initiatives supported by the U.S.
Department of Education, Department of Health and Human Services (HHS),
and Department of Justice that also conduct systematic reviews of
evaluation evidence to identify effective interventions.[Footnote 1] We
ascertained the procedures and criteria these federally supported
efforts used from interviews and document reviews.
We identified the types of interventions for which randomized
controlled experiments--the Coalition's primary evidence criterion--
are best suited and alternative methods for assessing effectiveness by
reviewing the program evaluation methodology literature and by having
our summaries of that literature reviewed by a diverse set of experts
in the field. We obtained reviews from seven experts who had published
on evaluation methodology, held leadership positions in the field, and
had experience in diverse subject areas and methodologies.
We conducted this performance audit from May 2008 through November 2009
in accordance with generally accepted government auditing standards.
Those standards require that we plan and perform the audit to obtain
sufficient, appropriate evidence to provide a reasonable basis for our
findings and conclusions based on our audit objectives. We believe that
the evidence obtained provides a reasonable basis for our findings and
conclusions based on our audit objectives.
Background:
Over the past two decades, several efforts have been launched to
improve federal government accountability and results, such as the
strategic plans and annual performance reports required under the
Government Performance and Results Act of 1993 (GPRA). The act was
designed to provide executive and congressional decision makers with
objective information on the relative effectiveness and efficiency of
federal programs and spending. In 2002, the Office of Management and
Budget (OMB) introduced the Program Assessment Rating Tool (PART) as a
key element of the budget and performance integration initiative under
President George W. Bush's governmentwide Management Agenda. PART is a
standard set of questions meant to serve as a diagnostic tool, drawing
on available program performance and evaluation information to form
conclusions about program benefits and recommend adjustments that may
improve results.
The success of these efforts has been constrained by lack of access to
credible evidence on program results. We previously reported that the
PART review process has stimulated agencies to increase their
evaluation capacity and available information on program results.
[Footnote 2] After 4 years of PART reviews, however, OMB rated 17
percent of 1,015 programs "results not demonstrated"--that is, did not
have acceptable performance goals or performance data. Many federal
programs, while tending to have limited evaluation resources, require
program evaluation studies, rather than performance measures, in order
to distinguish a program's effects from those of other influences on
outcomes.
Program evaluations are systematic studies that assess how well a
program is working, and they are individually tailored to address the
client's research question. Process (or implementation) evaluations
assess the extent to which a program is operating as intended. Outcome
evaluations assess the extent to which a program is achieving its
outcome-oriented objectives but may also examine program processes to
understand how outcomes are produced. When external factors such as
economic or environmental conditions are known to influence a program's
outcomes, an impact evaluation may be used in an attempt to measure a
program's net effect by comparing outcomes with an estimate of what
would have occurred in the absence of the program intervention. A
number of methodologies are available to estimate program impact,
including experimental and nonexperimental designs.
Concern about the quality of social program evaluation has led to calls
for greater use of randomized experiments--a method used more widely in
evaluations of medical than social science interventions. Randomized
controlled trials (or randomized experiments) compare the outcomes for
groups that were randomly assigned either to the treatment or to a
nonparticipating control group before the intervention, in an effort to
control for any systematic difference between the groups that could
account for a difference in their outcomes. A difference in these
groups' outcomes is believed to represent the program's impact. While
random assignment is considered a highly rigorous approach in assessing
program effectiveness, it is not the only rigorous research design
available and is not always feasible.
The Coalition for Evidence-Based Policy is a private, nonprofit
organization that was sponsored by the Council for Excellence in
Government from 2001 until the Council closed in 2009. The Coalition
aims to improve the effectiveness of social programs by encouraging
federal agencies to fund rigorous studies--particularly randomized
controlled trials--to identify effective interventions and to provide
strong incentives and assistance for federal funding recipients to
adopt such interventions.[Footnote 3] Coalition staff have advised OMB
and federal agencies on how to identify rigorous evaluations of program
effectiveness, and they manage a Web site called "Social Programs That
Work" that provides examples of evidence-based programs to "provide
policymakers and practitioners with clear, actionable information on
what works, as demonstrated in scientifically-valid studies...."
[Footnote 4]
In 2008, the Coalition launched a similar but more formal effort, the
Top Tier Evidence initiative, to identify only interventions that have
been shown in "well-designed and implemented randomized controlled
trials, preferably conducted in typical community settings, to produce
sizeable, sustained benefits to participants and/or society."[Footnote
5] At the same time, it introduced an advisory panel of evaluation
researchers and former government officials to make the final
determination. The Coalition has promoted the adoption of this
criterion in legislation to direct federal funds toward strategies
supported by rigorous evidence. By identifying interventions meeting
this criterion, the Top Tier Evidence initiative aims to assist
agencies, grantees, and others in implementing such provisions
effectively.
Federally Supported Initiatives to Identify Effective Interventions:
Because of the flexibility provided to recipients of many federal
grants, achieving these federal programs' goals relies heavily on
agencies' ability to influence their state and local program partners'
choice of activities. In the past decade, several public and private
efforts have been patterned after the evidence-based practice model in
medicine to summarize available effectiveness research on social
interventions to help managers and policymakers identify and adopt
effective practices. The Department of Education, HHS, and Department
of Justice support six initiatives similar to the Coalition's to
identify effective social interventions. These initiatives conduct
systematic searches for and review the quality of evaluations of
intervention effectiveness in a given field and have been operating for
several years.
We examined the processes used by these six ongoing federally supported
efforts to identify effective interventions in order to provide insight
into the choices of procedures and criteria that other independent
organizations made in attempting to achieve a similar outcome as the
Top Tier initiative: to identify interventions with rigorous evidence
of effectiveness. The Top Tier initiative, however, aims to identify
not all effective interventions but only those supported by the most
definitive evidence of effectiveness. The processes each of these
initiatives (including Top Tier) takes to identify effective
interventions are summarized in appendix I.
Evidence-Based Practice Centers:
In 1997, the Agency for Healthcare Research and Quality (AHRQ)
established the Evidence-based Practice Centers (EPC) (there are
currently 14) to provide evidence on the relative benefits and risks of
a wide variety of health care interventions to inform health care
decisions.[Footnote 6] EPCs perform comprehensive reviews and
synthesize scientific evidence to compare health treatments, including
pharmaceuticals, devices, and other types of interventions. The
reviews, with a priority on topics that impose high costs on the
Medicare, Medicaid, or State Children's Health Insurance (SCHIP)
programs, provide evidence about effectiveness and harms and point out
gaps in research. The reviews are intended to help clinicians and
patients choose the best tests and treatments and to help policy makers
make informed decisions about health care services and quality
improvement.[Footnote 7]
The Guide to Community Preventive Services:
HHS established the Guide to Community Preventive Services (the
Community Guide) in 1996 to provide evidence-based recommendations and
findings about public health interventions and policies to improve
health and promote safety. With the support of the Centers for Disease
Control and Prevention (CDC), the Community Guide synthesizes the
scientific literature to identify the effectiveness, economic
efficiency, and feasibility of program and policy interventions to
promote community health and prevent disease. The Task Force on
Community Preventive Services, an independent, nonfederal, volunteer
body of public health and prevention experts, guides the selection of
review topics and uses the evidence gathered to develop recommendations
to change risk behaviors, address environmental and ecosystem
challenges, and reduce disease, injury, and impairment. Intended users
include public health professionals, legislators and policy makers,
community-based organizations, health care service providers,
researchers, employers, and others who purchase health care services.
[Footnote 8]
HIV/AIDS Prevention Research Synthesis:
CDC established the HIV/AIDS Prevention Research Synthesis (PRS) in
1996 to review and summarize HIV behavioral prevention research
literature. PRS conducts systematic reviews to identify evidence-based
HIV behavioral interventions with proven efficacy in preventing the
acquisition or transmission of HIV infection (reducing HIV-related risk
behaviors, sexually transmitted diseases, HIV incidence, or promoting
protective behaviors). These reviews are intended to translate
scientific research into practice by providing a compendium of evidence-
based interventions to HIV prevention planners and providers and state
and local health departments for help with selecting interventions best
suited to the needs of the community.[Footnote 9]
Model Programs Guide:
The Office of Juvenile Justice and Delinquency Prevention established
the Model Programs Guide (MPG) in 2000 to identify effective programs
to prevent and reduce juvenile delinquency and related risk factors
such as substance abuse. MPG conducts reviews to identify effective
intervention and prevention programs on the following topics:
delinquency; violence; youth gang involvement; alcohol, tobacco, and
drug use; academic difficulties; family functioning; trauma exposure or
sexual activity and exploitation; and accompanying mental health
issues. MPG produces a database of intervention and prevention programs
intended for juvenile justice practitioners, program administrators,
and researchers.[Footnote 10]
National Registry of Evidence-Based Programs and Practices:
The Substance Abuse and Mental Health Services Administration (SAMHSA)
established the National Registry of Evidence-based Programs and
Practices (NREPP) in 1997 and provides the public with information
about the scientific basis and practicality of interventions that
prevent or treat mental health and substance abuse disorders.[Footnote
11] NREPP reviews interventions to identify those that promote mental
health and prevent or treat mental illness, substance use, or co-
occurring disorders among individuals, communities, or populations.
NREPP produces a database of interventions that can help practitioners
and community-based organizations identify and select interventions
that may address their particular needs and match their specific
capacities and resources.[Footnote 12]
What Works Clearinghouse:
The Institute of Education Sciences established the What Works
Clearinghouse (WWC) in 2002 to provide educators, policymakers,
researchers, and the public with a central source of scientific
evidence on what improves student outcomes. WWC reviews research on the
effectiveness of replicable educational interventions (programs,
products, practices, and policies) to improve student achievement in
areas such as mathematics, reading, early childhood education, English
language, and dropout prevention. The WWC Web site reports information
on the effectiveness of interventions through a searchable database and
summary reports on the scientific evidence.[Footnote 13]
Top Tier Initiative's Process Is Mostly Transparent:
The Coalition provides a clear public description on its Web site of
the first two phases of its process--search and selection to identify
candidate interventions. It primarily searches other evidence-based
practice Web sites and solicits nominations from experts and the
public. Staff post their selection criteria and a list of the
interventions and studies reviewed on their Web site. However, their
public materials have not been as transparent about the criteria and
process used in the second two phases of its process--review and
synthesize study results to determine whether an intervention met the
Top Tier criteria. Although the Coalition provides brief examples of
the panel's reasoning in making Top Tier selections, it has not fully
reported the panel's discussion of how to define sizable and sustained
effects in the absence of detailed guidance or the variation in
members' overall assessments of the interventions.
The Top Tier Initiative Clearly Described Its Process for Identifying
Interventions:
Through its Web site and e-mailed announcements, the Coalition has
clearly described how it identified interventions by searching the
strongest evidence category of 15 federal, state, and private Web sites
profiling evidence-based practices and by soliciting nominations from
federal agencies, researchers, and the general public. Its Web site
posting clearly indicated the initiative's search and selection
criteria: (1) early childhood interventions (for ages 0-6) in the first
phase of the initiative and interventions for children and youths (ages
7-18) in the second phase (starting in February 2009) and (2)
interventions showing positive results in well-designed and implemented
randomized experiments. Coalition staff then searched electronic
databases and consulted with researchers to identify any additional
randomized studies of the interventions selected for review. The July
2008 announcement of the initiative included its August 2007 "Checklist
for Reviewing a Randomized Controlled Trial of a Social Program or
Project, to Assess Whether It Produced Valid Evidence." The Checklist
describes the defining features of a well-designed and implemented
randomized experiment: equivalence of treatment and control groups
throughout the study, valid measurement and analysis, and full
reporting of outcomes. It also defines a strong body of evidence as
consisting of two or more randomized experiments or one large multi-
site study.
In the initial phase (July 2008 through February 2009), Coalition staff
screened studies of 46 early childhood interventions for design or
implementation flaws and provided the advisory panel with brief
summaries of the interventions and their results and reasons why they
screened out candidates they believed clearly did not meet the Top Tier
standard. Reasons for exclusion included small sample sizes, high
sample attrition (both during and after the intervention), follow-up
periods of less than 1 year, questionable outcome measures (for
example, teachers' reports of their students' behavior), and positive
effects that faded in later follow-up. Staff also excluded
interventions that lacked confirmation of effects in a well-implemented
randomized study. Coalition staff recommended three candidate
interventions from their screening review; advisory panel members added
two more for consideration after reviewing the staff summaries (neither
of which was accepted as top tier by the full panel). While the Top
Tier Initiative explains each of its screening decisions to program
developers privately, on its Web site it simply posts a list of the
interventions and studies reviewed, along with full descriptions of
interventions accepted as top tier and a brief discussion of a few
examples of the panel's reasoning.[Footnote 14]
Reviewers Defined the Top Tier Criteria through Case Discussion:
The Top Tier initiative's public materials are less transparent about
the process and criteria used to determine whether an intervention met
the Top Tier standard than about candidate selection. One panel member,
the lead reviewer, explicitly rates the quality of the evidence on each
candidate intervention using the Checklist and rating form. Coalition
staff members also use the Checklist to review the available evidence
and prepare detailed study reviews that identify any significant
limitations. The full advisory panel then discusses the available
evidence on the recommended candidates and holds a secret ballot on
whether an intervention meets the Top Tier standard, drawing on the
published research articles, the staff review, and the lead reviewer's
quality rating and Top Tier recommendation.
The advisory panel discussions did not generally dispute the lead
reviewer's study quality ratings (on quality of overall design, group
equivalence, outcome measures, and analysis reporting) but, instead,
focused on whether the body of evidence met the Top Tier standard (for
sizable, sustained effects on important outcomes in typical community
settings). The Checklist also includes two criteria or issues that were
not explicit in the initial statement of the Top Tier standard--whether
the body of evidence showed evidence of effects in more than one site
(replication) and provided no strong countervailing evidence. Because
neither the Checklist nor the rating form provides definitions of how
large a sizable effect should be, how long a sustained effect should
last, or what constituted an important outcome, the panel had to rely
on its professional judgment in making these assessments.
Although a sizable effect was usually defined as one passing tests of
statistical significance at the 0.05 level, panel members raised
questions about whether particular effects were sufficiently large to
have practical importance. The panel often turned to members with
subject matter expertise for advice on these matters. One member
cautioned against relying too heavily on the reported results of
statistical tests, because some studies, by conducting a very large
number of comparisons, appeared to violate the assumptions of those
tests and, thus, probably identified some differences between
experimental groups as statistically significant simply by chance.
The Checklist originally indicated a preference for data on long-term
outcomes obtained a year after the intervention ended, preferably
longer, noting that "longer-term effects...are of greatest policy and
practical importance."[Footnote 15] Panel members disagreed over
whether effects measured no later than the end of the second grade--at
the end of the intervention--were sufficiently sustained and important
to qualify as top tier, especially in the context of other studies that
tracked outcomes to age 15 or older. One panel member questioned
whether it was realistic to expect the effects of early childhood
programs to persist through high school, especially for low-cost
interventions; others noted that the study design did not meet the
standard because it did not collect data a year after the intervention
ended. In the end, a majority (but not all) of the panel accepted this
intervention as top tier because the study found that effects persisted
over all 3 program years, and they agreed to revise the language in the
Checklist accordingly.
Panel members disagreed on what constituted an important outcome. Two
noted a pattern of effects in one study on cognitive and academic tests
across ages 3, 5, 8, and 18. Another member did not consider cognitive
tests an important enough outcome and pointed out that the effects
diminished over time and did not lead to effects on other school-
related behavioral outcomes such as special education placement or
school drop-out. Another member thought it was unreasonable to expect
programs for very young children (ages 1-3) to show an effect on a
child at age 18, given all their other experiences in the intervening
years.
A concern related to judging importance was whether and how to
incorporate the cost of the intervention into the intervention
assessment. On one hand, there was no mention of cost in the Checklist
or intervention rating form. On the other hand, panel members
frequently raised the issue when considering whether they were
comfortable recommending the intervention to others. One aspect of this
was proportionality: they might accept an outcome of less policy
importance if the intervention was relatively inexpensive but would not
if it was expensive. Additionally, one panel member feared that an
expensive intervention that required a lot of training and monitoring
to produce results might be too difficult to successfully replicate in
more ordinary settings. In the February 2009 meeting, it was decided
that program cost should not be a criterion for Top Tier status but
should be considered and reported with the recommendation, if deemed
relevant.
The panel discussed whether a large multisite experiment should qualify
as evidence meeting the replication standard. One classroom-based
intervention was tested by randomly assigning 41 schools nationwide.
Because the unit of analysis was the school, results at individual
schools were not analyzed or reported separately but were aggregated to
form one experimental-control group comparison per outcome measure.
Some panel members considered this study a single randomized
experiment; others accepted it as serving the purpose of a replication,
because effects were observed over a large number of different
settings. In this case, limitations in the original study report added
to their uncertainty. Some panel members stated that if they had
learned that positive effects had been found in several schools rather
than in only a few odd cases, they would have been more comfortable
ruling this multisite experiment a replication.
Reviewers Initially Disagreed in Assessing Top Tier Status:
Because detailed guidance was lacking, panel members, relying on
individual judgment, arrived at split decisions (4-3 and 3-5) on two of
the first four early childhood interventions reviewed, and only one
intervention received a unanimous vote. Panel members expressed concern
that because some criteria were not specifically defined, they had to
use their professional judgment yet found that they interpreted the
terms somewhat differently. This problem may have been aggravated by
the fact that, as one member noted, they had not had a "perfect winner"
that met all the top tier criteria. Indeed, a couple of members
expressed their desire for a second category, like "promising," to
allow them to communicate their belief in an intervention's high
quality, despite the fact that its evidence did not meet all their
criteria. In a discussion of their narrow (4-3) vote at their next
meeting (February 2009), members suggested that they take more time to
discuss their decisions, set a requirement for a two-thirds majority
agreement, or ask for votes from members who did not attend the
meeting. The latter suggestion was countered with concern that absent
members would not be aware of their discussion, and the issue was
deferred to see whether these differences might be resolved with time
and discussion of other interventions. Disagreement over Top Tier
status was less a problem with later reviews, held in February and July
2009, when none of the votes on Top Tier status were split decisions
and three of seven votes were unanimous.
The Coalition reports that it plans to supplement guidance over time by
accumulating case decisions rather than developing more detailed
guidance on what constitutes sizable and sustained effects. The
December 2008 and May 2009 public releases of the results of the Top
Tier Evidence review of early childhood interventions provided brief
discussion of examples of the panel's reasoning for accepting or not
accepting specific interventions. In May 2009, the Coalition also
published a revised version of the Checklist that removed the
preference for outcomes measured a year after the intervention ended,
replacing it with a less specific reference: "over a long enough period
to determine whether the intervention's effects lasted at least a year,
hopefully longer."[Footnote 16]
At the February 2009 meeting, Coalition staff stated that they had
received a suggestion from external parties to consider introducing a
second category of "promising" interventions that did not meet the top
tier standard. Panel members agreed to discuss the idea further but
noted the need to provide clear criteria for this category as well. For
example, they said it was important to distinguish interventions that
lacked good quality evaluations (and thus had unknown effectiveness)
from those that simply lacked replication of sizable effects in a
second randomized study. It was noted that broadening the criteria to
include studies (and interventions) that the staff had previously
screened out may require additional staff effort and, thus, resources
beyond those of the current project.
Top Tier Follows Rigorous Standards but Is Limited for Identifying
Effective Interventions:
The Top Tier initiative's criteria for assessing evaluation quality
conform to general social science research standards, but other
features of the overall process differ from common practice for drawing
conclusions about intervention effectiveness from a body of research.
The initiative's choice of a broad topic fails to focus the review on
how to achieve a specific outcome. Its narrow evidence criteria yield
few recommendations and limited information on what works to inform
policy and practice decisions.
Review Initiatives Share Criteria for Assessing Research Quality:
The Top Tier and all six of the agency-supported review initiatives we
examined assess evaluation quality on standard dimensions to determine
whether a study provides credible evidence on effectiveness. These
dimensions include the quality of research design and execution, the
equivalence of treatment and comparison groups (as appropriate),
adequacy of samples, the validity and reliability of outcome measures,
and appropriateness of statistical analyses and reporting. Some
initiatives included additional criteria or gave greater emphasis to
some issues than others. The six agency-supported initiatives also
employed several features to ensure the reliability of their quality
assessments.
In general, assessing the quality of an impact evaluation's study
design and execution involves considering how well the selected
comparison protects against the risk of bias in estimating the
intervention's impact. For random assignment designs, this primarily
consists of examining whether the assignment process was truly random,
the experimental groups were equivalent before the intervention, and
the groups remained separate and otherwise equivalent throughout the
study. For other designs, the reviewer must examine the assignment
process even more closely to detect whether a potential source of bias
(such as higher motivation among volunteers) may have been introduced
that could account for any differences observed in outcomes between the
treatment and comparison groups. In addition to confirming the
equivalence of the experimental groups at baseline, several review
initiatives examine the extent of crossover or "contamination" between
experimental groups throughout the study because this could blur the
study's view of the intervention's true effects.
All seven review initiatives we examined assess whether a study's
sample size was large enough to detect effects of a meaningful size.
They also assess whether any sample attrition (or loss) over the course
of the study was severe enough to question how well the remaining
members represented the original sample or whether differential
attrition may have created significant new differences between the
experimental groups. Most review forms ask whether tests for
statistical significance of group differences accounted for key study
design features (for example, random assignment of groups rather than
individuals), as well as for any deviations from initial group
assignment (intention-to-treat analysis).[Footnote 17]
The rating forms vary in structure and detail across the initiatives.
For example, "appropriateness of statistical analyses" can be found
under the category "reporting of the intervention's effects" on one
form and in a category by itself on another form. In the Model Programs
Guide rating form, "internal validity"--or the degree to which observed
changes can be attributed to the intervention--is assessed through how
well both the research design and the measurement of program activities
and outcomes controlled for nine specific threats to validity.[Footnote
18] The EPC rating form notes whether study participants were blind to
the experimental groups they belonged to--standard practice in studies
for medical treatments but not as common in studies of social
interventions, while the PRS form does not directly address study
blinding in assessing extent of bias in forming study groups.
The major difference in rating study quality between the Top Tier
initiative and the six other initiatives is a product of the top tier
standard as set out in certain legislative provisions: the other
initiatives accept well-designed, well-conducted quasi-experimental
studies as credible evidence. Most of the federally supported
initiatives recognize well-conducted randomized experiments as
providing the most credible evidence of effectiveness by assigning them
their highest rating for quality of research design, but three do not
require them for interventions to receive their highest evidence
rating: EPC, the Community Guide, and National Registry of Evidence-
based Programs and Practices (NREPP). The Coalition has, since its
inception, promoted randomized experiments as the highest-quality,
unbiased method for assessing an intervention's true impact. Federal
officials provided a number of reasons for including well-conducted
quasi-experimental studies: (1) random assignment is not feasible for
many of the interventions they studied, (2) study credibility is
determined not by a particular research design but by its execution,
(3) evidence from carefully controlled experimental settings may not
reflect the benefits and harms observed in everyday practice, and (4)
too few high-quality, relevant random assignment studies were
available.
The Top Tier initiative states a preference for studies that test
interventions in typical community settings over those run under ideal
conditions but does not explicitly assess the quality (or fidelity) of
program implementation. The requirement that results be shown in two or
more randomized studies is an effort to demonstrate the applicability
of intervention effects to other settings. However, four other review
initiatives do explicitly assess intervention fidelity--the Community
Guide, MPG, NREPP, and PRS--through either describing in detail the
intervention's components or measuring participants' level of exposure.
Poor implementation fidelity can weaken a study's ability to detect an
intervention's potential effect and thus lessen confidence in the study
as a true test of the intervention model. EPC and the Community Guide
assess how well a study's selection of population and setting matched
those in which it is likely to be applied; any notable differences in
conditions would undermine the relevance or generalizability of study
results to what can be expected in future applications.
All seven initiatives have experienced researchers with methodological
and subject matter expertise rate the studies and use written guidance
or codebooks to help ensure ratings consistency. Codebooks varied but
most were more detailed than the Top Tier Checklist. Most of the
initiatives also provided training to ensure consistency of ratings
across reviewers. In each initiative, two or more reviewers rate the
studies independently and then reach consensus on their ratings in
consultation with other experts (such as consultants to or supervisors
of the review). After the Top Tier initiative's staff screening review,
staff and one advisory panel member independently review the quality of
experimental evidence available on an intervention, before the panel as
a group discussed and voted on whether it met the top tier standard.
However, because the panel members did not independently rate study
quality or the body of evidence, it is unknown how much of the
variation in their overall assessment of the interventions reflected
differences in their application of the criteria making up the Top Tier
standard.
Broad Scope Fails to Focus on Effectiveness in Achieving Specific
Outcomes:
The Top Tier initiative's topic selection, emphasis on long-term
effects, and narrow evidence criteria combine to provide limited
information on the effectiveness of approaches for achieving specific
outcomes. It is standard practice in research and evaluation syntheses
to pose a clearly defined research question--such as, Which
interventions have been found effective in achieving specific outcomes
of interest for a specific population?--and then assemble and summarize
the credible, relevant studies available to answer that question.
[Footnote 19] A well-specified research question clarifies the
objective of the research and guides the selection of eligibility
criteria for including studies in a systematic evidence review. In
addition, some critics of systematic reviews in health care recommend
using the intervention's theoretical framework or logic model to guide
analyses toward answering questions about how and why an intervention
works when it does.[Footnote 20] Evaluators often construct a logic
model--a diagram showing the links between key intervention components
and desired results--to explain the strategy or logic by which it is
expected to achieve its goals.[Footnote 21] The Top Tier initiative's
approach focuses on critically appraising and summarizing the evidence
without having first formulated a precise, unambiguous research
question and the chain of logic underlying the interventions'
hypothesized effects on the outcomes of interest.
Neither of the Top Tier initiative's topic selections--interventions
for children ages 0-6 or youths ages 7-18--identify either a particular
type of intervention, such as preschool or parent education, or a
desired outcome, such as healthy cognitive and social development or
prevention of substance abuse, that can frame and focus a review as in
the other effectiveness reviews. The other initiatives have a clear
purpose and focus: learning what has been effective in achieving a
specific outcome or set of outcomes (for example, reducing youth
involvement in criminal activity). Moreover, recognizing that an
intervention might be successful on one outcome but not another, EPC,
NREPP, and WWC rate the effectiveness of an intervention by each
outcome. Even EPC, whose scope is the broadest of the initiatives we
reviewed, focuses individual reviews by selecting a specific healthcare
topic through a formal process of soliciting and reviewing nominations
from key stakeholders, program partners, and the public. Their criteria
for selecting review topics include disease burden for the general
population or a priority population (such as children), controversy or
uncertainty over the topic, costs associated with the condition,
potential impact for improving health outcomes or reducing costs,
relevance to federal health care programs, and availability of evidence
and reasonably well-defined patient populations, interventions, and
outcome measures.
The Top Tier initiative's emphasis on identifying interventions with
long-term effects--up to 15 years later for some early childhood
interventions--also leads away from focusing on how to achieve a
specific outcome and could lead to capitalizing on chance results. A
search for interventions with "sustained effects on important life
outcomes," regardless of the content area, means assembling results on
whatever outcomes--special education placement, high school graduation,
teenage pregnancy, employment, or criminal arrest--the studies happen
to have measured. This is of concern because it is often not clear why
some long-term outcomes were studied for some interventions and not
others. Moreover, focusing on the achievement of long-term outcomes,
without regard to the achievement of logically related short-term
outcomes, raises questions about the meaning and reliability of those
purported long-term program effects. For example, without a logic model
or hypothesis linking preschool activities to improving children's self-
control or some other intermediate outcome, it is unclear why one would
expect to see effects on their delinquent behavior as adolescents.
Indeed, one advisory panel member raised questions about the mechanism
behind long-term effects measured on involvement in crime when effects
on more conventional (for example, academic) outcomes disappeared after
a few years. Later, he suggested that the panel should consider only
outcomes the researcher identified as primary. Coalition staff said
that reporting chance results is unlikely because the Top Tier criteria
require the replication of results in multiple (or multi-site) studies,
and they report any nonreplicated findings as needing confirmation in
another study.
Unlike efforts to synthesize evaluation results in some systematic
evidence reviews, the Top Tier initiative examines evidence on each
intervention independently, without reference to similar interventions
or, alternatively, to different interventions aimed at the same goal.
Indeed, of the initiatives we reviewed, only EPC and the Community
Guide directly compare the results of several similar interventions to
gain insight into the conditions under which an approach may be
successful. (WWC topic reports display effectiveness ratings by outcome
for all interventions they reviewed in a given content area, such as
early reading, but do not directly compare their approaches.) These two
initiatives explicitly aim to build knowledge about what works in an
area by developing logic models in advance to structure their
evaluation review by defining the specific populations and outcome
measures of interest. A third, MPG, considers the availability of a
logic model and the quality of an intervention's research base in
rating the quality of its evidence. Where appropriate evidence is
available, EPCs conduct comparative effectiveness studies that directly
compare the effectiveness, appropriateness, and safety of alternative
approaches (such as drugs or medical procedures) to achieving the same
health outcome. Officials at the other initiatives explained that they
did not compare or combine results from different interventions because
they did not find them similar enough to treat as replications of the
same approach. However, most initiatives post the results of their
reviews on their Web sites by key characteristics of the intervention
(for example, activities or setting), outcomes measured, and
population, so that viewers can search for particular types of
interventions or compare their results.
Narrow Evidence Criteria Yield Limited Guidance for Practitioners:
The Top Tier initiative's narrow primary criterion for study design
quality--randomized experiments only--diverges from the other
initiatives and limits the types of interventions they considered. In
addition, the exclusivity of its top tier standard also diverges from
the more common approach of rating the credibility of study findings
along a continuum and resulted in the panel's recommending only 6 of 63
interventions for ages 0-18 reviewed as providing "sizable, sustained
effects on important life outcomes." Thus, although they are not their
primary audience, the Top Tier initiative provides practitioners with
limited guidance on what works.
Two basic dimensions are assessed in effectiveness reviews: (1) the
credibility of the evidence on program impact provided by an individual
study or body of evidence, based on research quality and risk of bias
in the individual studies, and (2) the size and consistency of effects
observed in those studies. The six other evidence reviews report the
credibility of the evidence on the interventions' effectiveness in
terms of their level of confidence in the findings--either with a
numerical score (0 to 4, NREPP) or on a scale (high, moderate, low, or
insufficient, EPC). Scales permit an initiative to communicate
intermediate levels of confidence in an intervention's results and to
distinguish approaches with "promising" evidence from those with
clearly inadequate evidence. Federal officials from initiatives using
this more inclusive approach indicated that they believed that it
provides more useful information and a broader range of choices for
practitioners and policy makers who must decide which intervention is
most appropriate and feasible for their local setting and available
resources. To provide additional guidance to practitioners looking for
an intervention to adopt, NREPP explicitly rates the interventions'
readiness for dissemination by assessing the quality and availability
of implementation materials, resources for training and ongoing
support, and the quality assurance procedures the program developer
provides.
Some initiatives, like Top Tier, provide a single rating of the
effectiveness of an intervention by combining ratings of the
credibility and size (and consistency, if available) of intervention
effects. However, combining scores creates ambiguity in an intermediate
strength of evidence rating--it could mean that reviewers found strong
evidence of modest effects or weak evidence of strong effects. Other
initiatives report on the credibility of results and the effect sizes
separately. For example, WWC reports three summary ratings for an
intervention's result on each outcome measured: an improvement index,
providing a measure of the size of the intervention's effect; a rating
of effectiveness, summarizing both study quality and the size and
consistency of effects; and an extent of evidence rating, reflecting
the number and size of effectiveness studies reviewed. Thus, the viewer
can scan and compare ratings on all three indexes in a list of
interventions rank-ordered by the improvement index before examining
more detailed information about each intervention and its evidence of
effectiveness.
Randomized Experiments Can Provide the Most Credible Evidence of
Effectiveness under Certain Conditions:
In our review of the literature on program evaluation methods, we found
general agreement that well-conducted randomized experiments are best
suited for assessing intervention effectiveness where multiple causal
influences lead to uncertainty about what has caused observed results
but, also, that they are often difficult to carry out. Randomized
experiments are considered best suited for interventions in which
exposure to the intervention can be controlled and the treatment and
control groups' experiences remain separate, intact, and distinct
throughout the study. The evaluation methods literature also describes
a variety of issues to consider in planning an evaluation of a program
or of an intervention's effectiveness, including the expected use of
the evaluation, the nature and implementation of program activities,
and the resources available for the evaluation. Selecting a methodology
follows, first, a determination that an effectiveness evaluation is
warranted. It then requires balancing the need for sufficient rigor to
draw firm conclusions with practical considerations of resources and
the cooperation and protection of participants. Several other research
designs are generally considered good alternatives to randomized
experiments, especially when accompanied by specific features that help
strengthen conclusions by ruling out plausible alternative
explanations.
Conditions Necessary for Conducting Effectiveness Evaluations:
In reviewing the literature on evaluation research methods, we found
that randomized experiments are considered appropriate for assessing
intervention effectiveness only after an intervention has met minimal
requirements for an effectiveness evaluation--that the intervention is
important, clearly defined, and well-implemented and the evaluation
itself is adequately resourced. Conducting an impact evaluation of a
social intervention often requires the expenditure of significant
resources to both 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
reserved for when the effort and cost appear warranted. There may be
more interest in an impact evaluation when the intervention addresses
an important problem, there is interest in adopting the intervention
elsewhere, and preliminary evidence suggests its effects may be
positive, if uncertain. Of course, if the intervention's effectiveness
were known, then there would be no need for an evaluation. And if the
intervention was known or believed to be ineffective or harmful, then
it would seem wasteful as well as perhaps unethical to subject people
to such a test. In addition to federal regulations concerning the
protection of human research subjects, the ethical principles of
relevant professional organizations require evaluators to try to avoid
subjecting study participants to unreasonable risk, harm, or burden.
This includes obtaining their fully informed consent.[Footnote 22]
An impact evaluation is more likely to provide useful information about
what works when the intervention consists of clearly defined activities
and goals and has been well implemented. Having clarity about the
nature of intended activities and evidence that critical intervention
components were delivered to the intended targets helps strengthen
confidence that those activities caused the observed results; it also
improves the ability to replicate the results in another study.
Confirming that the intervention was carried out as designed helps rule
out a common explanation for why programs do not achieve their goals;
when done before collecting expensive outcome data, it can also avoid
wasting resources. Obtaining agreement with stakeholders on which
outcomes to consider in defining success also helps ensure that the
evaluation's results will be credible and useful to its intended
audience. While not required, having a well-articulated logic model can
help ensure shared expectations among stakeholders and define measures
of a program's progress toward its ultimate goals.
Regardless of the evaluation approach, an impact evaluation may not be
worth the effort unless the study is adequately staffed and funded to
ensure the study is carried out rigorously. If, for example, an
intervention's desired outcome consists of participants' actions back
on the job after receiving training, then it is critical that all
reasonable efforts are made to ensure that high-quality data on those
actions are collected from as many participants as possible.
Significant amounts of missing data raises the possibility that the
persons reached are different from those who were not reached (perhaps
more cooperative) and thus weakens confidence that the observed results
reflect the true effect of the intervention. Similarly, it is important
to invest in valid and reliable measures of desired outcomes to avoid
introducing error and imprecision that could blur the view of the
intervention's effect.
Interventions Where Random Assignment Is Well Suited:
We found in our review of the literature on evaluation research methods
that randomized experiments are considered best suited for assessing
intervention effectiveness where multiple causal influences lead to
uncertainty about program effects and it is possible, ethical, and
practical to conduct and maintain random assignment to minimize the
effect of those influences.
When Random Assignment Is Needed:
As noted earlier, when factors other than the intervention are expected
to influence change in the desired outcome, the evaluator cannot be
certain how much of any observed change reflects the effect of the
intervention, as opposed to what would have occurred anyway without it.
In contrast, controlled experiments are usually not needed to assess
the effects of simple, comparatively self-contained processes like
processing income tax returns. The volume and accuracy of tax returns
processed simply reflect the characteristics of the returns filed and
the agency's application of its rules and procedures. Thus, any change
in the accuracy of processed returns is likely to result from change in
the characteristics of either the returns or the agency's processes. In
contrast, an evaluation assessing the impact of job training on
participants' employment and earnings would need to control for other
major influences on those outcomes--features of the local job market
and the applicant pool. In this case, randomly assigning job training
applicants (within a local job market) to either participate in the
program (forming the treatment group) or not participate (forming the
control group) helps ensure that the treatment and control groups will
be equally affected.
When Random Assignment Is Possible, Ethical, and Practical:
Random assignment is, of course, suited only to interventions in which
the evaluator or program manager can control whether a person, group,
or other entity is enrolled in or exposed to the intervention. Control
over program exposure rules out the possibility that the process by
which experimental groups are formed (especially, self-selection) may
reflect preexisting differences between them that might also affect the
outcome variable and, thus, obscure the treatment effect. For example,
tobacco smokers who volunteer for a program to quit smoking are likely
to be more highly motivated than tobacco smokers who do not volunteer.
Thus, smoking cessation programs should randomly assign volunteers to
receive services and compare them to other volunteers who do not
receive services to avoid confounding the effects of the services with
the effects of volunteers' greater motivation.
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 the intervention in any case. This addresses one
concern about whether a control group experiment is ethical. In fact,
in many field settings, assignment by lottery has often been considered
the most equitable way to assign individuals to participate in programs
with limits on enrollment. Randomized experiments are especially well
suited to demonstration programs for which a new approach is tested in
a limited way before committing to apply it more broadly. Another
ethical concern is that the control group should not be harmed by
withholding needed services, but this can be averted by providing the
control group with whatever services are considered standard practice.
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.
Random assignment is also best suited for interventions in which the
treatment and control 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 that control group participants not access comparable
treatment in the community on their own (referred to as contamination).
Their doing so could blur the distinction between the two groups'
experiences. It is also preferred that control group and treatment
group members not communicate, because knowing that they are being
treated differently might influence their perceptions of their
experience and, thus, their behavior. Sometimes people selected for an
experimental treatment are motivated by the extra attention they
receive; sometimes those not selected are motivated to work harder to
compete with their peers. Thus, random assignment works best when
participants have no strong beliefs about the advantage of the
intervention being tested and information about their experimental
status is not publicly known. For example, in comparing alternative
reading curriculums in kindergarten classrooms, an evaluator needs to
ensure that the teachers are equally well trained and do not have
preexisting conceptions about the "better" curriculum. Sometimes this
is best achieved by assigning whole schools--rather than individuals or
classes--to the treatment and control groups, but this can become very
expensive, since appropriate statistical analyses now require about as
many schools to participate in a study as the number of classes
participating in the simpler design.
Interventions are well suited for random assignment if the desired
outcomes occur often enough to be observed with a reasonable sample
size or study length. Studies of infrequent but not rare outcomes--for
example, those occurring about 5 percent of the time--may require
moderately large samples (several hundred) to allow the detection of a
difference between the experimental and control groups. Because of the
practical difficulties of maintaining intact experimental groups over
time, randomized experiments are also best suited for assessing
outcomes that occur within 1 to 2 years after the intervention,
depending on the circumstances. Although an intervention's key desired
outcome may be a social, health, or environmental benefit that takes 10
or more years to fully develop, it may be prohibitively costly to
follow a large enough proportion of both experimental groups over that
time to ensure reliable results. Evaluators may then rely on
intermediate outcomes, such as high-school graduation, as an adequate
outcome measure rather than accepting the costs of directly measuring
long-term effects on adult employment and earnings.
Interventions for Which Random Assignment Is Not Well Suited:
Random assignment is not appropriate for a range of programs in which
one cannot meet the requirements that make this strategy effective.
They include entitlement programs or policies that apply to everyone,
interventions that involve exposure to negative events, or
interventions for which the evaluator cannot be sure about the nature
of differences between the treatment and control groups' experiences.
Random Assignment Is Not Possible:
For a few types of programs, random assignment to the intervention is
not possible. One is when all eligible individuals are exposed to the
intervention and legal restrictions do not permit excluding some people
in order to form a comparison group. This includes entitlement programs
such as veterans' benefits, Social Security, and Medicare, as well as
programs operating under laws and regulations that explicitly prohibit
(or require) a particular practice.
A second type of intervention for which random assignment is precluded
is broadcast media communication where the individual--rather than the
researcher--controls his or her exposure (consciously or not). This is
true of radio, television, billboard, and Internet programming, in
which the individual chooses whether and how long to hear or view a
message or communication. To evaluate the effect of advertising or
public service announcements in broadcast media, the evaluator is often
limited to simply measuring the audience's exposure to it. However,
sometimes it is possible to randomly assign advertisements to distinct
local media markets and then compare their effects to other similar but
distinct local markets.
A third type of program for which random assignment is generally not
possible is comprehensive social reforms consisting of collective,
coordinated actions by various parties in a community--whether school,
organization, or neighborhood. In these highly interactive initiatives,
it can be difficult to distinguish the activities and changes from the
settings in which they take place. For example, some community
development partnerships rely on increasing citizen involvement or
changing the relationships between public and private organizations in
order to foster conditions that are expected to improve services.
Although one might randomly assign communities to receive community
development support or not, the evaluator does not control who becomes
involved or what activities take place, so it is difficult to trace the
process that led to any observed effects.
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. Thus, one must wait for a hurricane or flood, for
example, to learn if efforts to strengthen buildings prevented serious
damage. Whether the evaluator is able to randomly apply different
approaches to strengthening buildings may depend on whether the
approaches appear to be equally likely to be successful in advance of a
test. In some cases, the possibility that the intervention may fail may
be considered an unacceptable risk. When evaluating alternative
treatments for criminal offenders, local law enforcement officers may
be unwilling to assign the offenders they consider to be the most
dangerous to the less restrictive treatments.
As implied by the previous discussion of when random assignment is well
suited, it may simply not be practical in a variety of circumstances.
It may not be possible to convince program staff to form control groups
by simple random assignment if it would deny services to some of the
neediest individuals while providing service to some of the less needy.
For example, individual tutoring in reading would usually be provided
only to students with the lowest reading scores. In other cases, the
desired outcome may be so rare or take so long to develop that the
required sample sizes or prospective tracking of cases over time would
be prohibitively expensive.
Finally, the evaluation literature cautions that as social
interventions become more complex, representing a diverse set of local
applications of a broad policy rather than a common set of activities,
randomized experiments may become less informative. When how much of
the intervention is actually delivered, or how it is expected to work,
is influenced by characteristics of the population or setting, one
cannot be sure about the nature of the difference between the treatment
and control group experiences or which factors influenced their
outcomes. Diversity in the nature of the intervention can occur at the
individual level, as when counselors draw on their experience to select
the approach they believe is most appropriate for each patient. Or it
can occur at a group level, as when grantees of federal flexible grant
programs focus on different subpopulations as they address the needs of
their local communities. In these cases, aggregating results over
substantial variability in what the intervention entails may end up
providing little guidance on what, exactly, works.
Rigorous Alternatives to Random Assignment Are Available:
In our review of the literature on evaluation research methods, we
identified several alternative methods for assessing intervention
effectiveness when random assignment is not considered appropriate--
quasi-experimental comparison group studies, statistical analyses of
observational data, and in-depth case studies. Although experts
differed in their opinion of how useful case studies are for estimating
program impacts, several other research designs are generally
considered good alternatives to randomized experiments, especially when
accompanied by specific features that help strengthen conclusions by
ruling out plausible alternative explanations.
Quasi-Experimental Comparison Groups:
Quasi-experimental comparison group designs resemble randomized
experiments in comparing the outcomes for treatment and control groups,
except that individuals are not assigned to those groups randomly.
Instead, unserved members of the targeted population are selected to
serve as a control group that resembles the treatment group as much as
possible on variables related to the desired outcome. This evaluation
design is used with partial coverage programs for which random
assignment is not possible, ethical, or practical. It is most
successful in providing credible estimates of program effectiveness
when the groups are formed in parallel ways and not based on self-
selection--for example, by having been turned away from an
oversubscribed service or living in a similar neighborhood where the
intervention is not available. This approach requires statistical
analyses to establish groups' equivalence at baseline.
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 are left otherwise comparable except for the intervention.
This technique is used where those considered most "deserving" are
assigned to treatment, in order to address ethical concerns about
denying services to those 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.
Statistical Analyses of Observational Data:
Interrupted time-series analysis compares trends in repeated measures
of an outcome for a group before and after an intervention or policy is
introduced, to learn if the desired change in outcome has occurred.
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 form or find an untreated comparison group, such as for
change in state laws defining alcohol impairment of motor vehicle
drivers ("blood alcohol concentration" laws). But because the technique
relies on the availability of comparable information about the past--
before a policy changed--it may be limited to use near the time of the
policy change. The need for lengthy data series means it is typically
used where the evaluator has access to long-term, detailed government
statistical series or institutional records.
Observational or cross-sectional studies first measure the target
population's level of exposure to the intervention rather than
controlling its exposure and then comparing the outcomes of individuals
receiving different levels of the intervention. Statistical analysis is
used to control for other plausible influences. Level of exposure to
the intervention can be measured by whether one was enrolled or how
often one participated or heard the program message. This approach is
used with full-coverage programs, for which it is impossible to
directly form treatment and control groups; nonuniform programs, in
which individuals receive different levels of exposure (such as to
broadcast media); 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
belts' effectiveness in preventing injuries during car crashes. Because
there is no evaluator control over assignment to the intervention, this
approach requires sophisticated statistical analyses to limit the
influence of any concurrent events or preexisting differences that may
be associated with why people had different exposure to the
intervention.
In-depth Case Studies:
Case studies have been recommended for assessing the effectiveness of
complex interventions in limited circumstances when other designs are
not available. In program evaluation, in-depth case studies are
typically used to provide descriptive information on how an
intervention operates and produces outcomes and, thus, may help
generate hypotheses about program effects. Case studies may also be
used to test a theory of change, as when the evaluator specifies in
advance the expected processes and outcomes, based on the program
theory or logic model, and then collects detailed observations
carefully designed to confirm or refute that model. This approach has
been recommended for assessing comprehensive reforms that are so deeply
integrated with the context (for example, the community) that no truly
adequate comparison case can be found.[Footnote 23] To support credible
conclusions about program effects, the evaluator must make specific,
refutable predictions of program effects and introduce controls for, or
provide strong arguments against, other plausible explanations for
observed effects. However, because a single case study most likely
cannot provide credible information on what would have happened in the
absence of the program, our experts noted that the evaluator cannot use
this design to reliably estimate the magnitude of a program's effect.
Features That Can Strengthen Any Effectiveness Evaluation:
Reviewing the literature and consulting with evaluation experts, we
identified additional measurement and design features that can help
strengthen conclusions about an intervention's impact from both
randomized and nonrandomized designs. In general, they involve
collecting additional data and targeting comparisons to help rule out
plausible alternative explanations of the observed results. Since all
evaluation methods have limitations, our confidence in concluding that
an intervention is effective is strengthened when the conclusion is
supported by multiple forms of evidence.
Collecting Additional Data:
Although collecting baseline data is an integral component of the
statistical approaches to assessing effectiveness discussed above, both
experiments and quasi-experiments would benefit from including pretest
measures on program outcomes as well as other key variables. First, by
chance, random assignment may not produce groups that are equivalent on
several important variables known to correlate with program outcomes,
so their baseline equivalence should always be checked. Second, in the
absence of random assignment, ensuring the equivalence of the treatment
and control groups on measures related to the desired outcome is
critical. The effects of potential self-selection bias or other
preexisting differences between the treatment and control groups can be
minimized through selection modeling or "propensity score analysis."
Essentially, one first develops a statistical model of the baseline
differences between the individuals in the treatment and comparison
groups on a number of important variables and then adjusts the observed
outcomes for the initial differences between the groups to identify the
net effect of the intervention.
Extending data collection either before or after the intervention can
help rule out the influence of unrelated historical trends on the
outcomes of interest. This is in principle similar to interrupted time-
series analysis, yielding more observations to allow analysis of trends
in outcomes over time in relation to the timing of program activities.
For example, one could examine whether the outcome measure began to
change before the intervention could plausibly have affected it, in
which case the change was probably influenced by some other factor.
Another way to attempt to rule out plausible alternative explanations
for observed results is to measure additional outcomes that are or are
not expected to be influenced by the treatment, based on program
theory. If one can predict a relatively unique pattern of expected
outcomes for the intervention, in contrast to an alternative
explanation, and if the study confirms that pattern, then the
alternative explanation becomes less plausible.
Targeting Comparisons:
In comparison group studies, the nature of the effect one detects is
defined by the nature of the differences between the experiences of the
treatment and control groups. For example, if the comparison group
receives no assistance at all in gaining employment, then the
evaluation can detect the full effect of all the employment assistance
(including child care) the treatment group receives. But if the
comparison group also receives child care, then the evaluation can
detect only the effect, or value added, of employment assistance above
and beyond the effect of child care. Thus, one can carefully design
comparisons to target specific questions or hypotheses about what is
responsible for the observed results and control for specific threats
to validity. For example, in evaluating the effects of providing new
parents of infants with health consultation and parent training at
home, the evaluator might compare them to another group of parents
receiving only routine health check-ups to control for the level of
attention the first group received and test the value added by the
parent training.
Sometimes the evaluator can capitalize on natural variations in
exposure to the intervention and analyze the patterns of effects to
learn more about what is producing change. For example, little or no
change in outcomes for dropouts--participants who left the program--
might reflect either the dropouts' lower levels of motivation compared
to other participants or their reduced exposure to the intervention.
But if differences in outcomes are associated with different levels of
exposure for administrative reasons (such as scheduling difficulties at
one site), then those differences may be more likely to result from the
intervention itself.
Gathering a Diverse Body of Evidence:
As reflected in all the review initiatives we identified for this
report, conclusions drawn from findings across multiple studies are
generally considered more convincing than those based on a single
study. The two basic reasons for this are that (1) each study is just
one example of many potential experiences with an intervention, which
may or may not represent that broader experience, and (2) each study
employs one particular set of methods to measure an intervention's
effect, which may be more or less likely than other methods to detect
an effect. Thus, an analysis that carefully considers the results of
diverse studies of an intervention is more likely to accurately
identify when and for whom an intervention is effective.
A recurring theme in the evaluation literature is the tradeoffs made in
constructing studies to rigorously identify program impact by reducing
the influence of external factors. Studies of interventions tested in
carefully controlled settings, a homogenous group of volunteer
participants, and a comparison group that receives no services at all
may not accurately portray the results that can be expected in more
typical operations. To obtain a comprehensive, realistic picture of
intervention effectiveness, reviewing the results of several studies
conducted in different settings and populations, or large multisite
studies, may help ensure that the results observed are likely to be
found, or replicated, elsewhere. This is particularly important when
the characteristics of settings, such as different state laws, are
expected to influence the effectiveness of a policy or practice applied
nationally. For example, states set limits on how much income a family
may have while receiving financial assistance, and these limits--which
vary considerably from state to state--strongly influence the
proportion of a state's assistance recipients who are currently
employed. Thus, any federal policy regarding the employment of
recipients is likely to affect one state's caseload quite differently
from that of another.
Because every research method has inherent limitations, it is often
advantageous to combine multiple measures or two or more designs in a
study or group of studies to obtain a more comprehensive picture of an
intervention. In addition to choosing whether to measure intermediate
or long-term outcomes, evaluators may choose to collect, for example,
student self-reports of violent behavior, teacher ratings of student
disruptive behavior, or records of school disciplinary actions or
referrals to the criminal justice system, which might yield different
results. While randomized experiments are considered best-suited for
assessing intervention impact, blended study designs can provide
supplemental information on other important considerations of policy
makers. For example, an in-depth case study of an intervention could be
added to develop a deeper understanding of its costs and implementation
requirements or to track participants' experiences to better understand
the intervention's logic model. Alternatively, a cross-sectional survey
of an intervention's participants and activities can help in assessing
the extent of its reach to important subpopulations.
Concluding Observations:
The Coalition provides a valuable service in encouraging government
adoption of interventions with evidence of effectiveness and in drawing
attention to the importance of evaluation quality in assessing that
evidence. Reliable assessments of the credibility of evaluation results
require expertise in research design and measurement, but their
reliability can be improved by providing detailed guidance and
training. The Top Tier initiative provides another useful model in that
it engages experienced evaluation experts to make these quality
assessments.
Requiring evidence from randomized experiments as sole proof of an
intervention's effectiveness is likely to exclude many potentially
effective and worthwhile practices for which random assignment is not
practical. The broad range of studies assessed by the six federally
supported initiatives we examined demonstrates that other research
designs can provide rigorous evidence of effectiveness if designed well
and implemented with a thorough understanding of their vulnerability to
potential sources of bias.
Assessing the importance of an intervention's outcomes entails drawing
a judgment from subject matter expertise--the evaluator must understand
the nature of the intervention, its expected effects, and the context
in which it operates. Defining the outcome measures of interest in
advance, in consultation with program stakeholders and other interested
audiences, may help ensure the credibility and usefulness of a review's
results. Deciding to adopt an intervention involves additional
considerations--cost, ease of use, suitability to the local community,
and available resources. Thus, practitioners will probably want
information on these factors and on effectiveness when choosing an
approach.
A comprehensive understanding of which practices or interventions are
most effective for achieving specific outcomes requires a synthesis of
credible evaluations that compares the costs and benefits of
alternative practices across populations and settings. The ability to
identify effective interventions would benefit from (1) better designed
and implemented evaluations, (2) more detailed reporting on both the
interventions and their evaluations, and (3) more evaluations that
directly compare alternative interventions.
Agency and Third-Party Comments:
The Coalition for Evidence-Based Policy provided written comments on a
draft of this report, reprinted in appendix II. The Coalition stated it
was pleased with the report's key findings on the transparency of its
process and its adherence to rigorous standards in assessing research
quality. While acknowledging the complementary value of well-conducted
nonrandomized studies as part of a research agenda, the Coalition
believes the report somewhat overstates the confidence one can place in
such studies alone. The Coalition and the Departments of Education and
Health and Human Services provided technical comments that were
incorporated as appropriate throughout the text. The Department of
Justice had no comments.
We are sending copies of this report to the Secretaries of Education,
Justice, and Health and Human Services; the Director of the Office of
Management and Budget; and appropriate congressional committees. The
report is also available at no charge on the GAO Web site at
[hyperlink, http://www.gao.gov].
If you have questions about this report, please contact me at (202) 512-
2700 or kingsburyn@gao.gov. Contacts for our offices of Congressional
Relations and Public Affairs are on the last page. Key contributors are
listed in appendix III.
Signed by:
Nancy Kingsbury, Ph.D.
Managing Director Applied Research and Methods:
[End of section]
Appendix I: Steps Seven Evidence-Based Initiatives Take to Identify
Effective Interventions:
1. Evidence-Based Practice Centers at the Agency for Healthcare
Research and Quality:
Search topic: Search for selected topics in health care services,
pharmaceuticals, and medical devices through:
* Electronic databases;
* Major journals;
* Conference proceedings;
* Consultation with experts.
Selected studies: Select:
* Randomized and quasi-experimental studies;
* Observational studies (e.g., cohort, case control).
Review studies quality: A technical panel of expert physicians, content
and methods experts, and other partners rates studies by outcome on:
* Study design and execution;
* Validity and reliability of outcome measures;
* Data analysis and reporting;
* Equivalence of comparison groups;
* Assessment of harm.
Synthesize evidence: Body of evidence on each outcome is scored on four
domains: risk of bias, consistency, directness, and precision of
effects. Strength of evidence for each outcome is classified as:
* High;
* Moderate;
* Low;
* Insufficient.
2. Guide to Community Preventive Services at the Centers for Disease
Control and Prevention:
Search topic: Search for selected population-based policies, programs,
and health care system interventions to improve health and promote
safety through:
* Electronic databases;
* Major journals;
* Conference proceedings;
* Consultation with experts.
Selected studies: Select:
* Randomized and quasi-experimental studies;
* Observational studies (e.g., time series, case control).
Review studies quality: In consultation with method and subject matter
experts, two trained reviewers independently rate studies using
standardized forms on:
* Study design and execution;
* Validity and reliability of outcome measures;
* Data analysis and reporting;
* Intervention fidelity;
* Selection of population and setting.
Synthesize evidence: Body of evidence is assessed on number of studies,
study quality, and size and consistency of effects to classify evidence
of effectiveness as:
* Strong;
* Sufficient;
* Insufficient.
3. HIV Prevention Research Synthesis at the Centers for Disease Control
and Prevention:
Search topic: Search for interventions that prevent new HIV/AIDS
infections or behaviors that increase the risk of infection through:
* Electronic databases;
* Major journals;
* Conference proceedings;
* Consultation with experts;
* Nominations solicited from the public.
Selected studies: Select randomized and quasi-experimental studies with
one or more positive outcomes.
Review studies quality: Pairs of trained reviewers—Ph.D.s or M.A.s in
behavioral science and health related areas—independently rate studies
using standardized forms and codebook on:
* Study design and execution;
* Validity and reliability of outcome measures;
* Data analysis and reporting;
* Equivalence of comparison groups;
* Assessment of harm.
Synthesize evidence: Ratings of study quality and strength of findings
are combined to classify interventions as:
* Best evidence;
* Promising evidence.
4. Model Programs Guide at the Office of Juvenile Justice and
Delinquency Prevention:
Search topic: Search for prevention and intervention programs to reduce
problem behaviors (juvenile delinquency, violence, substance abuse) in
at-risk juvenile population through:
* Electronic databases;
* Nominations solicited from the public.
Selected studies: Select randomized and quasi-experimental studies with
one or more positive outcomes and documentation of program
implementation (fidelity).
Review studies quality: A 3-person panel with 2 external Ph.D. content
area experts—with a codebook and consensual agreement—independently
rate studies on:
* Review studies quality: Study design and execution;
* Review studies quality: Validity and reliability of outcome measures;
* Review studies quality: Data analysis and reporting;
* Review studies quality: Equivalence of comparison groups;
* Review studies quality: Intervention fidelity;
* Review studies quality: Conceptual framework (logic and research
base).
Synthesize evidence: Ratings are combined across review criteria—
including consistency of evidence—to classify interventions as:
* Exemplary;
* Effective;
* Promising.
5. National Registry of Evidence-Based Programs and Practices at the
Substance Abuse and Mental Health Services Administration:
Search topic: Search for:
* Mental health promotion;
* Mental health treatment;
* Substance abuse prevention;
* Substance abuse treatment;
* Co-occurring disorders through:
- Electronic databases;
- Major journals;
- Nominations solicited from the public.
Selected studies: Select randomized and quasi-experimental studies with
one or more positive outcomes.
Review studies quality: Pairs of Ph.D. content specialists
independently rate studies on;
* Study design and execution;
* Validity and reliability of outcome measures;
* Data analysis and reporting;
* Intervention fidelity.
Pairs of providers and implementation experts independently rate
readiness for dissemination on:
* Implementation materials;
* Training and support resources;
* Quality assurance procedures.
Synthesize evidence: Summary research quality ratings (0–4) are
provided for statistically significant outcomes. Interventions
themselves are not rated. Scores on intervention readiness are averaged
to provide a score of 0–4.
6. Top Tier Evidence Initiative at the Coalition for Evidence-Based
Policy:
Search topic: Search for early childhood (ages 0–6) and youth (ages 7–
18) interventions through:
* Top evidence category of other evidence-based programs;
* Consultation with experts;
* Nominations solicited from the public;
Selected studies: Select randomized studies with one or more positive
outcomes.
Review studies quality: Team of M.A.s or Ph.D.s reviews studies and
selects candidates for the advisory panel’s review. Team reviews and
one advisory panel member rates studies on:
* Study design and execution;
* Validity and reliability of outcome measures;
* Data analysis and reporting;
* Equivalence of comparison groups.
Synthesize evidence: The advisory panel reviews studies and quality
ratings and assesses size and sustainability of effects in order to
classify as Top Tier.
7. What Works Clearinghouse at the Institute of Education Sciences:
Search topic: Search for interventions that improve student achievement
in:
* Early childhood education;
* Reading;
* Mathematics;
* Adolescent literacy;
* Dropout prevention;
* English language instruction through:
- Electronic databases;
- Major journals;
- Conference proceedings;
- Consultation with experts;
- Nominations solicited from the public.
Selected studies: Select randomized and quasi-experimental studies.
Review studies quality: Two Ph.D. research analysts independently rate
each study using codebook on:
* Study design and execution;
* Validity and reliability of outcome measures;
* Data analysis and reporting;
Ratings include:
* Meets evidence standards;
* Meets evidence standards with reservations.
Synthesize evidence: Across studies, ratings on quality of evidence and
effect’s direction, magnitude, and statistical significance for each
outcome are combined and classified as:
* Positive;
* Potentially positive;
* Mixed;
* None discernible;
* Potentially negative;
* Negative.
Number and size of studies are rated separately as:
* Small;
* Medium to large.
[End of section]
Appendix II: Comments from the Coalition for Evidence-Based Policy:
Coalition for Evidence-Based Policy:
900 19th Street, NW:
Suite 400:
Washington, DC 20006:
[hyperlink, http://www.coalition4evidence.org]
November 9, 2009:
Board of Advisors:
Robert Boruch:
University of Pennsylvania:
Jonathan Crane:
Coalition for Evidence-Based Policy:
David Ellwood:
Harvard University:
Judith Gueron:
MDRC:
Ron Haskins:
Brookings Institution:
Blair Hull:
Matlock Capital:
Robert Hoyt:
Jennison Associates:
David Kessler:
Former FDA Commissioner:
Jerry Lee:
Jerry Lee Foundation:
Dan Levy:
Harvard University:
Diane Ravitch:
New York University:
Howard Rolston:
Abt Associates:
Brookings institution:
Isabel Sawhill:
Brookings Institution:
Martin Seligman:
University of Pennsylvania:
Robert Solow:
Massachusetts Institute of Technology:
Nicholas Zill:
West., Inc.
President: Jon Baron:
jbaron@coalition4evidence.org:
202-380-3570:
The Coalition for Evidence-Based Policy is pleased with GAO's
confirmation of the Top Tier initiative's adherence to rigorous
standards and overall transparency:
The Coalition is pleased with the GAO report's key findings that the
Top Tier initiative's criteria conform to general social science
research standards (pp. 15-23), and that its process is mostly
transparent (pp. 9-15). We also agree with its observation that the Top
Tier initiative differs from common practice in its strong focus on
randomized experiments, and would add that this was the initiative's
goal from the start. Indeed, its stated purpose is to identify
interventions meeting the top tier standard set out in recent
Congressional legislation: "well-designed randomized controlled trials
(showing] sizeable, sustained effects on important...outcomes-(e.g.,
Public Laws 110-161 and 111-8).
Consistent with our initiative's unique focus on helping policymakers
distinguish the relatively few interventions meeting this top
evidentiary standard from the many that claim to, we have — as noted in
the GAO report — identified 6 interventions as Top Tier out of the 63
reviewed thus far. The value of this process to policymakers is
evidenced by the important impact these findings have already had on
federal officials and legislation. For example, the initiative's
findings for the Nurse-Family Partnership (NFP) have helped to spur the
Administration and Congress' proposed national expansion of evidence-
based home visitation. (The NFP study results arc cited in the
President's FY 2010 budget.) Similarly, the initiative's findings for
the Carrera Adolescent Pregnancy Prevention program and
Multidimensional Treatment Foster Care (MTFC) have helped inform the
Administration and Congress' proposed evidence-based teen pregnancy
prevention program. (The MTFC study results are cited in the Senate's
FY10 Labor-HHS-Education Appropriations Committee report.[Footnote 1])
In fact, 0M13 Director Peter Orszag recently posted on the OMB website
a summary of the Administration's "two-tiered approach" to home
visitation and teen pregnancy, which links to the Coalition's
website.[Footnote 2] The approach includes (i) funding for programs
backed by strong evidence, which he identifies as "the top tier;" and
(ii) additional funding for programs backed by "supportive evidence,"
with a requirement for rigorous evaluation that, if positive, could
move them into the top tier.
Consistent with this Administration approach, we recognize (and agree
with GAO) that nonrandomized studies provide important value — for
example, in (i) informing policy decisions in areas where well-
conducted randomized experiments are not feasible or not yet conducted;
ann (ii) identifying interventions that are particularly promising, and
therefore ready to be evaluated in more definitive randomized
experiments. We think the GAO report somewhat overstates the confidence
one can place in nonrandomized findings alone, per (i) a recent
National Academies recommendation[Footnote 3] that evidence of
effectiveness generally "cannot be considered definitive" without
ultimate confirmation in well-conducted randomized experiments, "even
if based on the next strongest designs;" and (ii) evidence that
findings from nonrandomized studies are often overturned in definitive
randomized experiments (see attachment). But the important and
complementary value of well-conducted nonrandomized studies as part of
an overall research agenda is a central theme of the Coalition's
approach to evidence-based policy reform.
In conclusion, we appreciate GAO's thoughtful analysis, and will use
its valuable observations to strengthen our initiative as it goes
forward. Although the Congressionally-established top tier standard
itself was not a main focus of the GAO report (as opposed to our
process), we have attached some brief background on the standard and
the reasons we support its use as an important clement of appropriate
policy initiatives.
Signed by:
Jon Baron, President:
[End of letter]
The Congressionally-established Top Tier evidence standard is based on
a well-established concept in the scientific community, and strong
evidence regarding the importance of random assignment:
Congress' Top Tier standard is based on a concept well-established in
the scientific community — that when results of multiple (or multi-
site) well-conducted randomized experiments, carried out in real-world
community settings, are available for a particular intervention, they
generally comprise the most definitive evidence regarding that
intervention's effectiveness. The standard further recognizes a key
concept articulated in a recent National Academies recommendation:
although many research methods can help identify effective
interventions, evidence of effectiveness generally "cannot be
considered definitive" without ultimate confirmation in well-conducted
randomized experiments, "even if based on the next strongest designs."
[Footnote 3]
Although promising findings in nonrandomized quasi-experimental studies
are valuable for decision making in the absence of stronger evidence,
too often such findings are overturned in subsequent, more definitive
randomized experiments. Reviews in medicine, for example, have found
that 50-80% of promising results from phase II (mostly quasi-
experimental) studies are overturned in subsequent phase III randomized
trials."[Footnote 4] Similarly, in education, eight of the nine major
randomized experiments sponsored by the Institute of Education Sciences
since its creation in 2002 have found weak or no positive effects for
the interventions being evaluated interventions which, in many cases,
were based on promising, mostly quasi-experimental evidence (e.g., the
LETRS teacher professional development program for reading
instruction).[Footnote 5] Systematic "design replication" studies
comparing well-conducted randomized experiments with quasi-experiments
in welfare, employment, and education policy have also found that many
widely-used and accepted quasi-experimental methods produce unreliable
estimates of program impact.[Footnote 6]
Thus, we support use of the Top Tier standard as a key element of
policy initiatives seeking to scale up interventions backed by the most
definitive evidence of sizeable, sustained effects, in areas where such
proven interventions already exist. The standard has a strong basis in
scientific authority and evidence, as reflected, for example, in the
recent National Academies recommendation.
References:
[1] Sen. Rept. 111-66.
[2] Peter Orszag's summary of the Administration's two-tiered approach
is posted at [hyperlink,
http://www.whitehouse.gov/omb/blog/09/06/08/BuildineRigorousEvidencetoDr
ivePolicy].
[3] National Research Council and Institute of Medicine. (2009).
Preventing Mental, Emotional, and Behavioral Disorders Among Young
People: Progress and Possibilities. Committee on Prevention of Mental
Disorders and Substance Abuse Among Children, Youth and Young Adults:
Research Advances and Promising Interventions. Mary Ellen O'Connell,
Thomas Boat, and Kenneth E. Warner, Editors. Board on Children, Youth,
and Families, Division of Behavioral and Social Sciences and Education.
Washington, DC: The National Academies Press. Recommendation 12-4, page
371.
[4] John P. A. Ioannidis, "Contradicted and Initially Stronger Effects
in Highly Cited Clinical Research,-Journal of the American Medical
Association, vol. 294, no. 2, July 13, 2005, pp. 218-228. Mohammad I.
Zia, Lillian L. Sin, Greg R. Pond, and Eric X. Chen, "Comparison of
Outcomes of Phase II Studies and Subsequent Randomized Control Studies
Using Identical Chemotherapeutic Regimens,".Journal of Clinical
Oncology, vol. 23, no. 28, October 1, 2005, pp. 6982-6991. John K. Chan
et. al., "Analysis of Phase II Studies on Targeted Agents and
Subsequent Phase III Trials: What Are the Predictors for Success,"
Journal of Clinical Oncology, vol. 26, no. 9, March 20, 2008.
[5] The Impact of Two Professional Development Interventions on Early
Reading Instruction and Achievement, Institute of Education Sciences,
NCEE 2008-4031, September 2008, [hyperlink,
http://ied.ed.gv/neee/pubs/20084030/].
[6] Howard S. Bloom, Charles Michalopoulos, and Carolyn J. Hill, "Using
Experiments to Assess Nonexperimental Comparison-Groups Methods for
Measuring Program Effects," in Learning More From Social Experiments:
Evolving Analytic Approaches, Russell Sage Foundation, 2005, pp. 173-
235. Thomas D. Cook, William R. Shadish, and Vivian C. Wong, "Three
Conditions Under Which Experiments and Observational Studies Often
Produce Comparable Causal Estimates: New Findings from Within-Study
Comparisons," Journal of Policy Analysis and Management, vol. 27, no.
4, pp. 724-50. Steve Glazerman, Dan M. Levy, and David Myers,
"Nonexperimental versus Experimental Estimates of Earnings Impact," The
American Annals of Political and Social Science, vol. 589, September
2003, pp. 63-93.
[End of section]
Appendix III: GAO Contact and Staff Acknowledgments:
GAO Contact:
Nancy Kingsbury, (202) 512-2700 or kingsburyn@gao.gov.
Staff Acknowledgments:
In addition to the person named above, Stephanie Shipman, Assistant
Director, and Valerie Caracelli made significant contributions to this
report.
[End of section]
Bibliography:
Agency for Healthcare Research and Quality. Systems to Rate the
Strength of Scientific Evidence: Summary. Evidence Report/Technology
Assessment No. 47. Rockville, Md.: U.S. Department of Health and Human
Services, March 2002. www.ahrq.gov/clinic/epcsums/strengthsum.htm.
Auspos, Patricia, and Anne C. Kubisch. Building Knowledge about
Community Change: Moving beyond Evaluations. New York: The Aspen
Institute, 2004.
Berk, Richard A. Randomized Experiments as the Bronze Standard.
California Center for Population Research On-Line Working Paper Series
CCPR-030-05. Los Angeles: August 2005.
http://repositories.cdlib.org/uclastat/papers/2005080201.
Boruch, Robert. “Encouraging the Flight of Error: Ethical Standards,
Evidence Standards, and Randomized Trials.” New Directions for
Evaluation no. 113 (spring 2007): 55–73.
Boruch, Robert F., and Ellen Foley. “The Honestly Experimental Society:
Sites and Other Entities as the Units of Allocation and Analysis in
Randomized Trials.” In Validity and Social Experimentation: Donald
Campbell’s Legacy, Leonard Bickman, ed. Thousand Oaks, Calif.: Sage,
2000.
Campbell, Donald T., and Julian C. Stanley. Experimental and Quasi-
experimental Designs for Research. Chicago: Rand McNally, 1966.
Chalmers, Iain. “Trying to Do More Good Than Harm in Policy and
Practice: The Role of Rigorous, Transparent, Up-to-date Evaluations.”
The Annals of the American Academy of Political and Social Science 589
(2003): 22–40.
Cook, Thomas D. “Randomized Experiments in Educational Policy Research:
A Critical Examination of the Reasons the Educational Evaluation
Community Has Offered for Not Doing Them.” Educational Evaluation and
Policy Analysis 24:3 (2002): 175–99.
European Evaluation Society. EES Statement: The Importance of a
Methodologically Diverse Approach to Impact Evaluation—Specifically
with Respect to Development Aid and Development Interventions. Nijkerk,
The Netherlands: December 2007. www.europeanevaluation.org.
Flay, Brian R., and others. “Standards of Evidence: Criteria for
Efficacy, Effectiveness, and Dissemination.” Prevention Science 6:3
(2005): 151–75.
Fulbright-Anderson, Anne C. Kibisch, and James P. Connell, eds. New
Approaches to Evaluating Community Initiatives. Vol. 2. Theory,
Measurement, and Analysis. Washington, D.C.: The Aspen Institute, 1998.
Glazerman, Steven, Dan M. Levy, and David Myers. “Nonexperimental
versus Experimental Estimates of Earnings Impacts.” The Annals of the
American Academy of Political and Social Science 589 (2003): 63–93.
Institute of Medicine. Knowing What Works in Health Care: A Roadmap for
the Nation. Washington, D.C.: The National Academies Press, 2008.
Jackson, N., and E. Waters. “Criteria for the Systematic Review of
Health Promotion and Public Health Interventions.” For the Guidelines
for Systematic Reviews in Health Promotion and Public Health Task
Force. Health Promotion International 20:4 (2005): 367–74.
Julnes, George, and Debra J. Rog. “Pragmatic Support for Policies on
Methodology.” New Directions for Evaluation no. 113 (spring 2007): 129–
47.
Mark, Melvin M., and Charles S. Reichardt. “Quasi-experimental and
Correlational Designs: Methods for the Real World When Random
Assignment Isn’t Feasible.” In The Sage Handbook of Methods in Social
Psychology, Carol Sansone, Carolyn C. Morf, and A. T. Panter, eds.
Thousand Oaks, Calif.: Sage, 2004.
Moffitt, Robert A. “The Role of Randomized Field Trials in Social
Science Research: A Perspective from Evaluations of Reforms of Social
Welfare Programs.” The American Behavioral Scientist 47:5 (2004): 506–
40.
National Research Council, Center for Education. Scientific Research in
Education. Washington, D.C.: National Academies Press, 2002.
National Research Council, Committee on Law and Justice. Improving
Evaluation of Anticrime Programs. Washington, D.C.: National Academies
Press, 2005.
Orr, Larry L. Social Experiments: Evaluating Public Programs with
Experimental Methods. Thousand Oaks, Calif.: Sage, 1999.
Posavac, Emil J., and Raymond G. Carey. Program Evaluation: Methods and
Case Studies, 6th ed. Upper Saddle River, N.J.: Prentice Hall, 2003.
Rossi, Peter H., Mark W. Lipsey, and Howard E. Freeman. Evaluation: A
Systematic Approach, 7th ed. Thousand Oaks, Calif.: Sage, 2004.
Trochim, William M. K. President, American Evaluation Association,
Chair, AEA Evaluation Policy Task Force. Letter to Robert Shea,
Associate Director for Administration and Government Performance,
Office of Management and Budget, Washington, D.C., March 7, 2008, and
attachment, “Comments on ‘What Constitutes Strong Evidence of a Program’
s Effectiveness?’” www.eval.org/EPTF.asp
Victora, Cesar G., Jean-Pierre Habicht, and Jennifer Bryce. “Evidence-
Based Public Health: Moving beyond Randomized Trials.” American Journal
of Public Health 94:3 (March 2004): 400–05.
West, Stephen, and others. “Alternatives to the Randomized Controlled
Trial.” American Journal of Public Health 98:8 (August 2008): 1359–66.
[End of section]
Related GAO Products:
Juvenile Justice: Technical Assistance and Better Defined Evaluation
Plans Will Help to Improve Girls’ Delinquency Programs. GAO-09-721R.
Washington, D.C.: July 24, 2009.
Health-Care-Associated Infections in Hospitals: Leadership Needed from
HHS to Prioritize Prevention Practices and Improve Data on These
Infections. GAO-08-283. Washington, D.C.: March 31, 2008.
School Mental Health: Role of the Substance Abuse and Mental Health
Services Administration and Factors Affecting Service Provision. GAO-08-
19R. Washington, D.C.: October 5, 2007.
Abstinence Education: Efforts to Assess the Accuracy and Effectiveness
of Federally Funded Programs. GAO-07-87. Washington, D.C.: October 3,
2006.
Program Evaluation: OMB’s PART Reviews Increased Agencies’ Attention to
Improving Evidence of Program Results. GAO-06-67. Washington, D.C.:
October 28, 2005.
Program Evaluation: Strategies for Assessing How Information
Dissemination Contributes to Agency Goals. GAO-02-923. Washington,
D.C.: September 30, 2002.
The Evaluation Synthesis. GAO/PEMD-10.1.2. Washington, D.C.: March
1992.
Designing Evaluations. GAO/PEMD-10.1.4. Washington, D.C.: March 1991.
Case Study Evaluations. GAO/PEMD-10.1.9. Washington, D.C.: November
1990.
[End of section]
Footnotes:
[1] In addition, the federal Interagency Working Group on Youth
Programs Web site [hyperlink, http://www.findyouthinfo.gov] provides
interactive tools and other resources to help youth-serving
organizations assess community assets, identify local and federal
resources, and search for evidence-based youth programs.
[2] GAO, Program Evaluation: OMB’s PART Reviews Increased Agencies’
Attention to Improving Evidence of Program Results, [hyperlink,
http://www.gao.gov/products/GAO-06-67] (Washington, D.C.: October 28,
2005), p. 28.
[3] See Coalition for Evidence-Based Policy, [hyperlink,
http://www.coalition4evidence.org].
[4] See Coalition for Evidence-Based Policy, Social Programs That Work,
[hyperlink, http://www.evidencebasedprograms.org].
[5] See Coalition for Evidence-Based Policy, Top Tier Evidence,
[hyperlink, http://toptierevidence.org]. The criterion is also
sometimes phrased more simply as interventions that have been shown in
well-designed randomized controlled trials to produce sizable,
sustained effects on important outcomes.
[6] AHRQ was formerly called the Agency for Health Care Policy and
Research.
[7] See Agency for Healthcare Research and Quality, Effective Health
Care, [hyperlink, http://www.effectivehealthcare.ahrq.gov].
[8] See Guide to Community Preventive Services, [hyperlink,
http://www.thecommunityguide.org/index.html].
[9] See Centers for Disease Control and Prevention, HIV/AIDS Prevention
Research Synthesis Project, [hyperlink,
http://www.cdc.gov/hiv/topics/research/prs].
[10] See Office of Juvenile Justice and Delinquency Prevention
Programs, OJJDP Model Programs Guide, [hyperlink,
http://www2.dsgonline.com/mpg].
[11] It was established as the National Registry of Effective
Prevention Programs; it was expanded in 2004 to include mental health
and renamed the National Registry of Evidence-based Programs and
Practices.
[12] See NREPP, SAMHSA’s National Registry of Evidence-based Programs
and Practices, [hyperlink, http://www.nrepp.samhsa.gov].
[13] See IES What Works Clearinghouse, [hyperlink,
http://ies.ed.gov/ncee/wwc].
[14] See [hyperlink, http://toptierevidence.org].
[15] Coalition for Evidence-Based Policy, “Checklist for Reviewing a
Randomized Controlled Trial of a Social Program or Project, to Assess
Whether It Produced Valid Evidence,” August 2007, p. 5. [hyperlink,
http://toptierevidence.org].
[16] Coalition, 2007, p. 5.
[17] In intention-to-treat analysis, members of the treatment and
control groups are retained in the group to which they were originally
assigned, even if some treatment group members failed to participate in
or complete the intervention or some control group members later gained
access to the intervention. See Checklist, p. 4.
[18] These factors were initially outlined in the classic research
design book by Donald T. Campbell and Julian C. Stanley, Experimental
and Quasi-Experimental Designs for Research (Chicago: Rand McNally,
1963).
[19] GAO, The Evaluation Synthesis, [hyperlink,
http://www.gao.gov/products/GAO/PEMD-10.1.2] (Washington, D.C.: March
1992); Institute of Medicine, Knowing What Works in Health Care
(Washington, D.C.: National Academies Press, 2008); Iain Chalmers,
“Trying to Do More Good Than Harm in Policy and Practice: The Role of
Rigorous, Transparent, Up-to-Date Evaluations,” The Annals of the
American Academy of Political and Social Science (Thousand Oaks,
Calif.: Sage, 2003); Agency for Healthcare Research and Quality,
Systems to Rate the Strength of Scientific Evidence (Rockville, Md.:
2002).
[20] Institute of Medicine, Knowing What Works; N. Jackson and E.
Waters, “Criteria for the Systematic Review of Health Promotion and
Public Health Interventions,” Health Promotion International (2005):
367–74.
[21] GAO, Program Evaluation: Strategies for Assessing How Information
Dissemination Contributes to Agency Goals, [hyperlink,
http://www.gao.gov/products/GAO-02-923] (Washington, D.C.: Sept. 30,
2002).
[22] See 45 C.F.R. Part 46 (2005) and, for example, the American
Evaluation Association’s Guiding Principles for Evaluators, revised in
2004. [hyperlink,
http://www.eval.org/Publications/GuidingPrinciples.asp].
[23] See Karen Fulbright-Anderson, Anne S. Kubisch, and James P.
Connell, eds., New Approaches to Evaluating Community Initiatives, vol.
2, Theory, Measurement, and Analysis (Washington, D.C.: Aspen
Institute, 1998), and Patricia Auspos and Anne S. Kubisch, Building
Knowledge about Community Change: Moving Beyond Evaluations
(Washington, D.C.: Aspen Institute, 2004).
[End of section]
GAO's Mission:
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.
Obtaining Copies of GAO Reports and Testimony:
The fastest and easiest way to obtain copies of GAO documents at no
cost is through GAO's Web site [hyperlink, http://www.gao.gov]. Each
weekday, GAO posts newly released reports, testimony, and
correspondence on its Web site. To have GAO e-mail you a list of newly
posted products every afternoon, go to [hyperlink, http://www.gao.gov]
and select "E-mail Updates."
Order by Phone:
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 Web site,
[hyperlink, 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.
To Report Fraud, Waste, and Abuse in Federal Programs:
Contact:
Web site: [hyperlink, http://www.gao.gov/fraudnet/fraudnet.htm]:
E-mail: fraudnet@gao.gov:
Automated answering system: (800) 424-5454 or (202) 512-7470:
Congressional Relations:
Ralph Dawn, Managing Director, dawnr@gao.gov:
(202) 512-4400:
U.S. Government Accountability Office:
441 G Street NW, Room 7125:
Washington, D.C. 20548:
Public Affairs:
Chuck Young, Managing Director, youngc1@gao.gov:
(202) 512-4800:
U.S. Government Accountability Office:
441 G Street NW, Room 7149:
Washington, D.C. 20548: