How Cognitive Biases Can Undermine Program Scale-Up ...
-
Upload
khangminh22 -
Category
Documents
-
view
2 -
download
0
Transcript of How Cognitive Biases Can Undermine Program Scale-Up ...
How Cognitive Biases Can Undermine Program Scale-Up Decisions
Susan Mayer, Rohen Shah, and Ariel Kalil
University of Chicago
August 2020
Forthcoming in List, J., Suskind, D., & Supplee, L. H. (Ed.). The Scale-up Effect in Early Childhood and Public Policy: Why interventions lose impact at scale and what we can do about it. Routledge.
HOW COGNITIVE BIASES CAN UNDERMINE SCALE UP
2
Abstract
Many social programs that yield positive results in pilot studies fail to yield such results when
implemented at scale. A variety of possible explanations aim to account for this failure. In this
chapter we emphasize the potential role of cognitive biases in the decision-making process that
takes an intervention from an idea to a fully implemented program. In that process, researchers,
practitioners, program participants, and funders all make decisions that influence whether and
how the program is implemented and evaluated and how evaluations are interpreted. Each
decision is subject to the cognitive biases of the decision maker and these biases can increase the
chance that false positive evaluation results lead to scaled—and in this case, failed—
implementation. Although the potential for cognitive biases to influence the scale-up process is
large, the research evidence is nearly nonexistent, suggesting the need for a psychologically
informed understanding of the scale-up process.
HOW COGNITIVE BIASES CAN UNDERMINE SCALE UP
3
Introduction
Many social programs that yield positive results in pilot studies fail to work when scaled
to a large population. A seemingly successful pilot intervention might fail to scale for any
number of reasons, including unrepresentative samples of participants, low implementation
fidelity, treatment heterogeneity, and other factors that are discussed in this book. But problems
may also arise because of the cognitive biases of those involved in the process of initiating,
evaluating, and scaling a program.
Researchers, practitioners, program participants, and policy makers must all make
decisions that influence whether a program is scaled up, and each decision is potentially subject
to cognitive bias (Sunstein, 2014). In this chapter we focus on three cognitive biases that
previous research has shown have an important influence on consequential decisions, namely
confirmation bias, status quo or default bias, and bandwagon bias.
A cognitive bias is a consistent and predictable propensity to make decisions in a
particular way that deviates from common principles of logic and probability. As such, cognitive
biases sometimes result in suboptimal decisions. Psychologists generally believe that cognitive
biases fulfill a current or evolutionary psychological need that is “hardwired” in the brain and
thus form the default patterns of human cognition (e.g. Haselton, Nettle, & Murray 2016;
Kahneman, 2011). Cognitive biases arise because when individuals are faced with a decision,
they confront vast quantities of information, time constraints, a social context and history that
provides meaning to the information, and limited brain power to process all the information. Under
these circumstances, humans use shortcuts that may produce quick and satisfying decisions
characterized by perceptual distortion, inaccurate judgment, and inconsistent interpretations,
collectively referred to as irrational (e.g. Kahneman & Tversky, 1972, 1974; Ariely, 2008;
HOW COGNITIVE BIASES CAN UNDERMINE SCALE UP
4
Gilovich, Griffin, & Kahneman, 2002). For example, Tversky and Kahneman (1974), who have
done pioneering work on cognitive biases, demonstrated that human probability judgments are
consistently insensitive to knowledge of prior probability and sample size but are just as
consistently influenced by irrelevant factors, such as the ease of imagining an event or the
provision of an unrelated random number. Other cognitive biases are related to social influences
on decision making, including the need for conformity and the desire to preserve one’s identity
(Akerlof & Kranton, 2010).
Not all social scientists agree about how many cognitive biases exist and many of the
cognitive biases discussed in the research literature overlap. However, there is broad agreement
that cognitive biases should be distinguished from errors in judgement, computational errors, or
other errors that are the result of misinformation. Errors that people make due to misinformation
can be corrected by providing accurate information, but cognitive biases are “hardwired”—they
are difficult to change and cannot be changed by simply providing information. Changing the
“choice architecture” (Thaler & Sunstein, 2008), which is the structure in which decisions are
made, and using behavioral tools, such as commitment devices and reminders (Mayer, Kalil,
Oreopoulos, & Gallegos, 2019; Kalil, Mayer, & Gallegos, 2020), can help people adapt to
cognitive biases in specific contexts. However, it is unclear whether people can completely
overcome these biases (see Nisbett, 2015 and Hershfield, 2011 for the view that at least some
biases can be overcome). Cognitive biases should also be distinguished from responses to
incentives that might result in distortions to decision making. Responding to an incentive is
neither surprising nor contrary to logic or probability assessment, even if the incentive itself
encourages behavior that is suboptimal from the point of view of the person offering the
incentive.
HOW COGNITIVE BIASES CAN UNDERMINE SCALE UP
5
Not everyone shares the same view of the origin, persistence, or influence of cognitive
biases (e.g., see Gigerenzer, 1996 for a critique of “prospect theory,” which was among the first
formal theories on decision-making to account for cognitive biases). About 185 cognitive biases
have been identified, but some are trivial and others rest on dubious empirical research. In
addition, not all decisions are seriously compromised by cognitive biases and, in some cases,
cognitive biases are beneficial. For instance, cognitive shortcuts mitigate against “decision
fatigue,” i.e., the decrease in decision-making quality that results from a prolonged period of
decision making (Baumeister, 2003).
Until recently, economists and other social scientists seldom accounted for consistent
deviations from “rationality” in how people gather and interpret information in models of
decision making. However, a growing research literature and the popularization of “nudge” units
integrating behavioral insights in governments around the world has led to a new awareness of
the ways in which cognitive biases influence policy decisions (Afif, Islan, Calvo-Gonzalez, &
Dalton, 2019).
This chapter focuses on cognitive biases that are related to the information that policy
makers and others involved in decisions to scale up programs seek out and consume. We focus
on how these biases might lead to scaling up programs that should not have been scaled up in the
first place. In doing so, we suggest a possible role for cognitive biases in explaining the
commonly observed “voltage drop” in scaled up programs and hence provide insight into the
question “why do so many programs fail to scale?” The discussion of the potential influence of
cognitive biases in the scale-up process is necessarily speculative because there is almost no
research on this topic. The intent of this chapter is to make the case that we need clear research
on this topic.
HOW COGNITIVE BIASES CAN UNDERMINE SCALE UP
6
How Cognitive Biases Might Influence Scaling Decisions
Ideally, the process leading to a social program being implemented at scale begins with a
theory about what treatment will change the behavior of participants in a positive way. The
theory then gets translated into a program with a set of procedures that, when followed, might
result in a change in the behavior of program participants. If funders are willing, the program is
then implemented in a pilot or at a small scale to see if behavior changes in the predicted way. If
researchers decide that a pilot program or replication study changes behavior in the predicted
way, and if funders are again willing, the program may then be implemented at a larger scale.
In this ideal process, researchers make objective decisions about the research based only
on uncovering the true effect of the intervention. However, researchers are not always
completely objective. Social scientific research often requires researchers to make decisions
about both the research process and the interpretation of results. Put another way, researchers
have many degrees of freedom (Simmons, Nelson, & Simonsohn, 2011) or “forking paths”
(Gelman & Loken, 2014) in the research process that require decisions. Such decisions include
the sample size, whether to omit some observations, what factors to include in a model, how to
construct outcome measures, and so on. These decisions can influence the type of data collected
from a pilot study and the way that results get analyzed, ultimately shaping the conclusions that
researchers draw about whether or not a program is effective.
When researchers face ambiguous decisions in the research process, they tend to make
decisions that increase the chance of achieving statistically significant results (p ≤ .05) for the
desired outcome (Babcock & Loewenstein, 1997; Dawson, Gilovich, & Regan, 2002; Gilovich,
1983; Hastorf & Cantril, 1954; Kunda, 1990; Zuckerman, 1979). This is a threat to accurate
HOW COGNITIVE BIASES CAN UNDERMINE SCALE UP
7
inference from the results of policy research (Al-Ubaydli, List, & Suskind 2019; Al-Ubaydli,
Lee, List, Mackevicius, & Suskind, 2019). Researcher decisions are influenced by the incentives
built into the research process: positive results are more highly rewarded through publication and
funding. Other influences on decisions in the research process arise from the funding source.
This potential influence is widely recognized in academia, as indicated by the requirement that
researchers report conflicts of interest for funded research. Industry-funded clinical trials and
cost-effectiveness analyses, for instance, yield positive results far more often than studies that are
not funded or conducted by industry or other interested entities (Davidson, 1986; Rochon et al.,
1994; Cho & Bero, 1996; Friedberg, Saffran, Stinson, Nelson, & Bennett, 1999; Krimsky 2013).
The many degrees of freedom in the research process have two important impacts on the
decision to scale up a program. First, it means that conventional measures of statistical
significance do not necessarily provide a strong signal in favor of scientific claims (Cohen, 1994;
Ioannidis, 2005; Simmons et al., 2011; Francis, 2013; Button et al., 2013; Gelman & Carlin
2014) because researchers may have made decisions that were designed to produce those results.
Relying solely on statistical significance of an estimate can lead to accepting false positive
results (e.g., Simmons et al., 2011; Maniadis, Tufano, and List 2014). When researchers accept
false positive results for program impacts, scaling those programs can lead to disappointing null
effects in a scaled-up implementation.
Second, the fact that there are many ambiguous decision points in the research process
opens the door for cognitive biases to influence how researchers design, implement, and interpret
the results of pilot studies. Moreover, researchers are not the only ones involved in the decision
to scale up an intervention. Policy makers, practitioners, and funders also play important roles in
HOW COGNITIVE BIASES CAN UNDERMINE SCALE UP
8
that decision and the fact that they, too, have many degrees of freedom in their decision process
leaves open the possibility for their cognitive biases to influence the decision.
A pilot study is intended to influence individuals’ belief about whether a program is truly
effective. According to the economic model of scaling, when faced with the results of a pilot
study, individuals should update their prior belief about an intervention’s efficacy by calculating
a post-study probability (PSP) that the intervention is effective. To do this, they should begin
with their belief about the probability of effectiveness prior to the study, calculate the likelihood
that a pilot study’s results are true based on the study’s level of power and statistical significance
and then up-date their belief about the probability that a program is effective. The power of a
study is the likelihood of detecting a treatment effect if the effect truly exists. Power tends to be
greater when the size of the sample in an intervention is greater. However, larger samples result
in smaller effects being statistically significant at conventional levels, so both power and
statistical significance are aids for interpreting the credibility of effect sizes and they do not tell
us alone whether a program is worth scaling.
Al-Ubaydli, Lee, List, Mackevicius, and Suskind (2019) recommend not scaling an
intervention until the PSP for the intervention reaches 0.95 or greater. They also recommend
starting with a low prior belief about an intervention’s effectiveness—0.10 on a scale of 0 to 1—
when there is minimal information about a program’s effectiveness. Given these
recommendations, it would take two well-powered studies with positive, statistically significant
results to reach the PSP recommended for scaling.
With each statistically significant finding, the PSP increases (Maniadis et al., 2014).
However, the PSP also should be adjusted for studies finding insignificant results. An
insignificant pilot result should lower the PSP relative to the prior belief.
HOW COGNITIVE BIASES CAN UNDERMINE SCALE UP
9
This formal process of updating prior beliefs assumes an unbiased decision maker. But
researchers, like everyone else, are subject to cognitive biases that affect their judgements. These
biases can increase the chance of their accepting false positive results and scaling programs that
are likely to fail. In the next three sections, we describe how three different cognitive biases
might influence the decision to scale a program:
• Confirmation Bias: The tendency to search for and interpret information in a way that
confirms one's preconceptions.
• Status Quo Bias: The tendency to want things to stay relatively the same.
• Bandwagon Bias: The tendency to do or believe things solely because many other people
do or believe it.
Confirmation Bias
As already noted, updating one’s beliefs about the efficacy of a program in light of the
results of a pilot is fundamental to the decision about whether to scale the program. In this
section, we describe how confirmation bias can alter both research procedures and the
interpretation of results in ways that can result in an individual accepting false positive results.
Confirmation bias leads people to gather, interpret, and recall information in a way that
conforms to their previously existing beliefs (Jonas, Schulz-Hardt, Frey, & Thelen, 2001;
Wason, 1960, 1968; Kleck &Wheaton 1967). Imagine Jack and Jill. Jack supports candidate A
and Jill supports candidate B. Jack seeks out news stories and social media contacts who support
A while Jill does the same for B. Both of their views are confirmed by the evidence that they
find. Similarly, when Jack sees evidence in support of candidate B, he is likely to find reasons
not to believe the evidence and when Jill sees evidence in support of A, she is likely to discount
that, too. This same process of selectively attending to information occurs in many arenas.
HOW COGNITIVE BIASES CAN UNDERMINE SCALE UP
10
In his iconic test of confirmation bias, Peter Wason (1960) developed the Rule Discovery
Test. He gave subjects three numbers and asked them to come up with the rule that applied to the
selection of those numbers. Subjects could ask questions and devise other sets of numbers that
researchers would say did or did not conform to the rule. In this way, subjects were asked to
determine if their rule was correct. Given the sequence of “2-4-6", subjects typically formed a
hypothesis that the rule was even numbers. They tested this by presenting other sequences of
even numbers, all of which were correct. After several correct tries, the subjects believed that
they had discovered the rule. In fact, they had failed, because the rule was “increasing numbers.”
The significance of this study is that almost all subjects formed a hypothesis and tested number
sequences that conformed only to their hypothesis. Very few tried a number sequence that might
disprove their hypothesis. Few subjects asked questions to try to falsify their hypothesis.
Wason’s Rule Discovery Test demonstrated that most people do not try at all to test their
hypotheses critically but instead try only to confirm their hypothesis.
Since this pioneering experiment, dozens of both laboratory and field experiments have
demonstrated the existence and importance of confirmation bias. For example, one study found
that when presented with an article containing ambiguous information on the pros and cons of
policies, conservatives and liberals alike interpreted the same, neutral article as making a case for
their own prior viewpoint (Angner, 2016). A similar phenomenon might occur when a pilot
program measuring several equally important outcomes produces results that show that it
benefits some outcomes but not others. In these circumstances, the presence of confirmation bias
suggests that a person with a prior belief in the program might interpret equivocal results such as
these as supporting the program being scaled. Alternatively, suppose a researcher has a strong
prior belief that an intervention will be effective. She reads about the results of two pilot studies
HOW COGNITIVE BIASES CAN UNDERMINE SCALE UP
11
that are consistent with her prior. As noted earlier, this should be enough evidence to make her
PSP above the 0.95 threshold if it is the only evidence available. But what if there were also
eight other similarly powered studies that all produced nonsignificant results? A researcher
facing some studies with positive results and others with negative results should update her
beliefs by increasing her prior twice for the two positive results, and then decreasing it eight
times for the negative results. The resulting PSP will likely fall far short of the 0.95 threshold.
The influence of confirmation bias suggests that she will give greater weight to the two results
that agree with her prior belief than to the eight that disagree with it. That is, the resulting PSP
the researcher calculates will be above 0.95 and meet the recommendations in the economic
model of scaling, even though the true PSP taking into account the eight negative results should
be far lower. This could result in her deciding to scale up the program that should not be scaled
up.
While there is not much research on how confirmation bias may influence the research
process, some evidence shows that it occurs in the peer review process (Mahoney 1977;
Goldsmith, Blalock, Bobkova, & Hall, 2006; Lee et al., 2013). In an early study on this topic,
Mahoney (1977) asked 75 journal reviewers to review manuscripts that had identical
introductory sections and experimental procedures but either reported data that were consistent
or were inconsistent with the reviewer’s theoretical perspective. The results demonstrated that
reviewers were strongly biased against manuscripts that reported results that were contrary to
their own theoretical perspective. In other words, they were more inclined to publish results that
conformed to their prior belief than results that disconfirmed it.
Another way that confirmation bias could influence the decision to scale a program is
through funding decisions. A substantial research literature in political science shows that
HOW COGNITIVE BIASES CAN UNDERMINE SCALE UP
12
government agencies have agendas, are influenced by fashions and their own paradigms that
create strong prior beliefs about what will work, and fund programs based on these prior beliefs
(e.g. see Vaganay 2016 for a summary). Similarly, foundations have missions and agendas that
result in their selectively choosing research to support those pre-existing beliefs. This cognitive
bias can lead to funding weakly vetted programs that have a high probability of failure.
Some kinds of data that are frequently used to evaluate programs may also be subject to
confirmation bias and therefore lead to scaling programs that are likely to fail. For example, in
program evaluations, teacher reports are often used to measure changes in student outcomes (e.g.
Brackett, Rivers, Reyes, & Salovey, 2012). Parent responses are also frequently used as an
outcome measure in evaluations of parent programs because what parents do in the home is often
unobservable (see e.g. York & Loeb 2014). This kind of self-reported data is problematic for
many reasons, but one important reason is the possibility of confirmation bias among
practitioners and participants. For example, several studies show that fidelity to a program is
greater when practitioners believe the program will work and that they are more likely to believe
a program will work when it aligns with their prior beliefs (Mihalic, Fagan, & Argamaso, 2008;
Ringwalt et al., 2003; Rohrbach, Graham, & Hansen, 1993). We can speculate that parents or
teachers are more likely to believe that there has been a positive change in their behavior when
they participate in a program that is consistent with their priors about what works than when they
participate in a program that is inconsistent with what they believe works. This belief may not
correspond with actual changes in behavior leading researchers to scale up a program that is
likely to fail.
Finally, confirmation bias could result in researchers and others placing more weight on
the results of early pilot programs than on later data about the same program (Baron, 2008;
HOW COGNITIVE BIASES CAN UNDERMINE SCALE UP
13
Allahverdyan and Galstyan 2014). This tendency is known as the “recency effect.” If early
results confirm one’s priors, it becomes more likely that latter disconfirming evidence will be
discounted. This too can result in relying on false positive results to scale a program.
Status Quo (Default) Bias
It has long been recognized that governments are slow to adopt new programs even when
evidence shows that the new program generates benefits that far outweigh its cost. A variety of
explanations for this phenomenon have been offered (e.g., Fernandez and Rodrik, 1991),
including status quo or default bias. Status quo bias refers to the fact that individuals tend to
prefer the current state of events to a change, even if a change is in one’s interest (Samuelson &
Zeckhauser, 1988; Kahneman, Knetsch, & Thaler, 1991). There are two main explanations for
status quo bias. The first suggests that people are hardwired to assume that the status quo is
good, or at least better than an alternative (Eidelman & Crandall, 2014). The second suggests that
it is a result of loss aversion (Moshinsky & Bar-Hillel, 2010; Quattrone & Tversky, 1988). Loss
aversion is a cognitive bias that refers to the documented phenomenon that the psychological
discomfort of a loss is about twice as powerful as the psychological pleasure of an equal gain
(Kahneman & Tversky, 1979). Because loss is so painful, people prefer the known status quo
over changes that might result in loss.
Status quo bias can result in people choosing the default option. A classic demonstration
of status quo bias is in decisions about retirement programs. Research has shown that 401(k)
participation increases dramatically when companies switch from an enrollment system whereby
employees must opt into the program to an enrollment system in which employees are
automatically enrolled and must opt out (Madrian & Shea, 2001; Choi, Laibson, Madrian, &
Metrick, 2004). This demonstrates that employees prefer whatever the current state or status quo
HOW COGNITIVE BIASES CAN UNDERMINE SCALE UP
14
is over having to make a decision to change something—even when the change is in their own
financial interest.
In terms of scaling up a program, imagine a researcher who has completed a pilot study
of a reading curriculum. The results show that the new curriculum raises reading skill by an
impressive amount compared to a control group receiving the curriculum currently used in
schools. The new curriculum is also easier and less expensive to implement. As Chambers and
Norton argue in Chapter 14 of this book, a scenario like this suggests that de-implementation of
the current curriculum should be considered. So, a researcher might then try to convince school
districts to replace the old curriculum with the new one. Anyone who has been in this position
knows that school districts are unlikely to jump at the chance to implement the new curriculum
(see e.g. Berkovick, 2011; Terhart, 2013). A possible explanation for this reluctance is status quo
bias. Given the choice of a new superior program to a current program, the choice is often the
current program, even when the cost of transition to the new program makes it cost effective (see
e.g. Polites & Karahanna, 2012).
Status quo bias has been demonstrated in many experimental situations. For example, in
one study, participants were given the option to reduce their anxiety while waiting for an electric
shock. When doing nothing was the status quo option, participants frequently did not select the
option that would reduce their anxiety (Suri, Sheppes, Schwartz & Gross, 2013).
Outside the laboratory, status quo bias has been used to explain many choices that seem
irrational or suboptimal. Consider the fact that drivers are often offered the choice between
expensive car insurance that protects the insured’s right to sue and cheaper plans that restrict the
right to sue. In a study of insurance decisions, one state had set the default plan as the expensive
one whereas the cheaper plan was the default in another state. In both states, individuals were
HOW COGNITIVE BIASES CAN UNDERMINE SCALE UP
15
more likely to choose the default option (Johnson, Hersey, Meszaros & Kunnreuth, 1993). Many
studies have examined the effects of changing from opt-in to opt-out choice structures, on
everything from charitable giving (Sanders, Halpern & Service 2013), to end-of-life care
(Halpern et al., 2013), and choosing green energy (Pichert & Katsikopoulos 2008), among
others. Status quo bias has also been demonstrated in research on the advantage of incumbents in
elections and on the preference of retirees for their current investment distribution over a new
distribution, even when the new distribution would improve their rate of return (Beshears, Choi,
Laibson, & Madrian, 2006; Ameriks & Zeldes, 2000).
The fact that status quo bias appears to have a widespread influence in consequential
policy decisions raises the possibility of its influencing decisions in the scale-up process. We can
speculate that status quo bias plays a role in why a new program that works may not be scaled
up. If a decision maker must choose between a piloted program with positive results that would
displace an existing program, status quo bias gives weight to a decision in favor of the existing
program, all else equal. On the other hand, status quo bias can also help explain why a new
program that does not work may get scaled. If a decision maker faces a choice between a pilot
program that is already being implemented versus an entirely new program, status quo bias may
lead them to stick with and scale-up the existing program—even if its results have been
disappointing. In short, status quo bias might prevent a potentially beneficial program from being
scaled or result in scaling a program that is not beneficial.
Bandwagon Bias
Social scientists have long tried to understand how some things become fashionable
compared to other things that appear to be just as functional and cost less. The attraction of
luxury-brand items like Louis Vuitton and Gucci bags, Cartier and Piaget watches, or BMW and
HOW COGNITIVE BIASES CAN UNDERMINE SCALE UP
16
Mercedes cars seems not to align with rational decision making. Similarly, it is hard to explain
the popularity of pet rocks or videos that go viral. And why do some diet and exercise regimes
become fads while others, including those with more evidence of effectiveness, fail to become
widespread?
In 1951, Solomon Asch developed a laboratory experiment intended to understand this
kind of behavior. He recruited students to participate in what they thought was a vision
experiment. Researchers put these students in classrooms with other students who were actually
a part of the experimental design, posing as other participant-students. Everyone in the room was
shown a picture with three lines of varying lengths with some obviously longer than others. Each
person in the room was asked to state out loud which was the longest line. In one experimental
condition, the students who were part of the experimental design all identified the wrong line. On
average, over a third of the other participant students went along with the clearly incorrect
choice. In some trials, as many as 75 percent of participant students went along with the clearly
wrong students. In a control condition with no students placed in the room as part of the
experimental design, virtually all students chose the correct answer.
Asch’s experiment demonstrated the existence of bandwagon bias. Bandwagon bias
refers to the tendency of people to adopt what others do or think simply because everyone else is
doing it. The more people who adopt a particular trend, the more likely it becomes that other
people will also “hop on the bandwagon” (see e.g, Henshel & Johnson, 1987). Bandwagon bias
is one of several biases that arise from peer or other social influences. Psychologists believe that
bandwagon bias, like other cognitive biases, aids in processing information for quick decision
making. But in this case, the information is processed in light of the views and behavior of
HOW COGNITIVE BIASES CAN UNDERMINE SCALE UP
17
others. Since Asch’s experiment, a large research literature has demonstrated the importance of
bandwagon bias to human decision making.
Bandwagon bias can interfere with scale-up decisions because it can magnify the impact
of false positives. For reasons described above, false positives are more likely to occur than the
5% implied by a result that is significant at 𝑝 = .05. When a pilot intervention with a false
positive result is promoted by a trusted or influential (but not necessarily expert) source, or is
generously funded, it can result in others duplicating the program before additional research is
done. As noted above, initial pilot results might be favored over later evidence due to
confirmation bias. When early but erroneous results are favored by influential sources, it can
create a bandwagon bias that magnifies the confirmation bias. Once a critical mass has hopped
“on the bandwagon” it becomes harder to convince people to change their now-strong prior
beliefs. This can lead to scaling up a program that is likely to fail.
While fashion and toy choices are clear examples of bandwagon bias, research also
suggests that this cognitive bias occurs in more serious and costly decisions. Bandwagon bias has
led to the widespread adoption of programs in education with weak or no basis in evidence other
than the popularity of the program. These education fads include open classrooms, whole
language learning, and “new math” (Phillips, 2015). A large literature documents bandwagon
biases in medicine, including the use of estrogen to prevent heart disease (Rossouw 1996), and
other medical procedures with dubious efficacy (McCormack & Greenhalgh, 2000; Trotter,
2002). In these cases, bandwagon bias may have resulted in more harm than good both because
the policies may have been harmful but also because bandwagon ideas can crowd out policies or
program that are more likely to be productive.
Factors That Can Amplify or Reduce the Influence of Cognitive Biases
HOW COGNITIVE BIASES CAN UNDERMINE SCALE UP
18
We speculate that additional factors in the decision-making environment and the type of
available evidence increase the chance that cognitive biases influence the scale up-decision. We
know of no research that has tested how these factors interact with cognitive biases. However,
we believe that they should be considered in future research. Understanding what factors
exacerbate the influence of cognitive biases would help us find ways to ameliorate that influence.
The decision-making environment can greatly affect the extent to which cognitive biases
influence decisions about scaling up a program. Decision-making environments that
encourage reflective judgment and an understanding of the potential influence of cognitive biases
can reduce their influence. For example, recent research indicates that delaying a decision by
even a 10th of a second significantly increases decision accuracy (Teichert, Ferrera & Grinband,
2014). As noted earlier, meeting the 0.95 PSP threshold before scaling a program implies that the
decision to scale should be based on the results of multiple independent replications in part
because the longer time period required to make the decision reduces the likelihood that
cognitive bias affects the final decision.
When decision makers have a lot of autonomy, cognitive biases are likely to have more
influence than when there are many checks in the decision-making process. As we described
above, researchers have a great deal of autonomy in making decisions about the research process.
Checks on those decisions are supposed to occur in the peer review process, including
conference and seminar presentations. However, these potentially valuable checks may
sometimes also be influenced by participants’ cognitive biases. Autonomy among other decision
makers also raises the likelihood for the influence of cognitive biases. In schools and many social
programs, practitioners have a great deal of autonomy in decisions about what programs get
implemented and in the day-to-day operation of programs. This autonomy may result in scaling
HOW COGNITIVE BIASES CAN UNDERMINE SCALE UP
19
programs that should not be scaled because it allows practitioners to make decisions influenced
by cognitive biases.
While multiple checks in the decision process may reduce cognitive biases they may also
result in decision-makers’ becoming more entrenched in their views. When people have invested
a lot of effort to develop a program, implement a pilot and defend the results in seminars and
other places they tend to want to protect that investment by not changing their mind about
whether the program works. This can foster confirmation bias as well as status quo bias.
Common advice to avoid confirmation bias is to establish systems that allow for reviews of
decisions, to use many sources of information, and other strategies that lengthen the decision-
making process. However, while these procedures might allow for more input, the additional
investment may also mean that the decision is more entrenched in the end. Striking the balance
between creating decision structures that have sufficient reflection and review but are not so
burdensome as to entrench ideas could be an important goal in reforming the research process.
The type of available evidence can also exacerbate or ameliorate the influence of
cognitive biases (Baddeley, Curtis, & Wood, 2004). Although cognitive biases influence
decisions in a way that is “irrational,” high-quality information may moderate the extent of that
influence. The amount and quality of available confirming evidence and the amount and quality
of available disconfirming evidence (or the ratio of confirming to disconfirming evidence) can
influence the extent of confirmation bias, status quo bias, and bandwagon bias among decision-
makers. If both confirming and disconfirming evidence is readily available, the chance for
cognitive bias to influence decisions about scaling a program is high because any view can be
supported by the evidence. If there are, for example, five studies with positive results and five
with null (or negative) results, a person with a high prior belief in the program will favor the
HOW COGNITIVE BIASES CAN UNDERMINE SCALE UP
20
positive results and discount the null results. As more negative results accumulate relative to the
positive results, it becomes harder to discount the negative results and cherry pick the positive
results. Different program domains may have a different evidence base, thus leading to more or
less confirmation bias in programs in different domains. In terms of status quo bias, additional
consistent information can reduce the probability that a decision will lead to a psychological loss
(Iyengar & Lepper 2000; Dean, Ozgur, & Yesufcan, 2014). In terms of bandwagon bias, having
both confirming and disconfirming evidence equally available implies that neither is dominant,
and there is therefore no “bandwagon” to jump on. Consequently, a way to minimize the
influence of cognitive bias is to build a strong and believable research base on the efficacy of
programs.
We can also build opportunities into the decision-making process to truly consider
alternatives to help mitigate against cherry-picking evidence from biased searches. Because it is
hard to ignore high-quality and consistent evidence about a program, we should support repeated
pilots of important promising programs and give equal weight in the publication process to
positive, negative, and insignificant results of well-powered studies. This aligns well with the
recommendation from the economic model of scaling to reward researchers for replicating
results as well as designing interventions that can be independently replicated. Doing so would
financially de-incentivize a researcher from only looking for evidence that confirms their priors.
Finally, although incentives are not cognitive biases, they may result in biases such as
confirmation bias. We should remove pernicious incentives from the research process. The
economic model of scaling shows us that publication bias results in a higher likelihood of scaling
up false positives, and changing incentives where null results are published can help mitigate
HOW COGNITIVE BIASES CAN UNDERMINE SCALE UP
21
this. This balanced increase in available evidence will help ensure that decision-makers
accurately adjust their PSP.
Conclusion
There is not enough data to know the extent to which cognitive biases influence the
decision to scale up an intervention. However, research shows that across domains, even decision
makers who believe themselves to be objective and dispassionate can be subject to cognitive
biases because they are not conscious of them. Even when they are, they may have a hard time
overcoming them without changes to the environment in which decisions are made. Furthermore,
research has shown that in many policy domains, cognitive biases play an important role in
decision making and that they can reduce the quality of decision making. This suggests that a
psychologically informed understanding of the scale-up decision process that includes the roles
of both researchers and non-researchers may provide evidence that improves decision making
and reduces the voltage drop effect. This research should include both laboratory and field
experiments to understand the influence of theoretically probable cognitive biases on decisions in
the research process, the funding process, the peer review process, and the implementation
process, in order to identify where in those processes cognitive biases are resulting in suboptimal
decision making.
In sum, confirmation bias occurs when one looks for information that supports one’s
existing beliefs and rejects data that goes against those beliefs; status quo bias arises from our
tendency to want things to stay relatively the same; and bandwagon bias occurs because humans
have a tendency to do or believe things solely because many other people do or believe it. From
a practical perspective, it is important in the decision-making process not only to acknowledge
that these biases exist but also to build in processes to mitigate against them. As noted, building
HOW COGNITIVE BIASES CAN UNDERMINE SCALE UP
22
decision-making environments that encourage reflective judgment and an understanding of the
potential influence of cognitive biases can reduce their influence. Crucially, it is important for
the decision-maker to look for ways to challenge what she thinks she sees. She could accomplish
this by surrounding herself with a diverse group of people and actively soliciting dissenting
views. A team of decision-makers can be designed to include one or more individuals who
challenges others’ opinions. Or, team members can be assigned to play the role of "devil's
advocate" for major decisions or to role-play situations that take into account multiple
perspectives. Whether in the research or policy process, adopting these decision-check
techniques at critical junctures in the decision-making process hold promise for reaching more
well-rounded and less biased decisions and avoiding the pitfalls that can arise when cognitive
bias interferes with rational decision making.
HOW COGNITIVE BIASES CAN UNDERMINE SCALE UP
23
References Afif, Z., Islan, W. W., Calvo-Gonzalez, O., & Dalton, A. G. (2019). Behavioral Science Around
the World: Profiles of 10 Countries (English). eMBeD brief. Washington, D.C.: World
Bank Group.
Akerlof, G., & Kranton, R. (2010). Identity economics. Princeton University Press.
Allahverdyan A.E. and Galstyan A. (2014). Opinion Dynamics with Confirmation Bias. PLoS
ONE 9(7): e99557. doi:10.1371/journal.pone.0099557
Al-Ubaydli, O., List, J. A., & Suskind, D. (2019). The Science of Using Science : Towards an
Understanding of the Threats to Scaling Experiments. National Bureau of Economic
Research.
Ameriks, J., & Zeldes, S. P. (2000). How do household portfolio shares vary with age?. Working
paper: Columbia University and TIAA-CREF.
Angner, E. (2016). A course in behavioral economics (2nd ed.). Macmillan International Higher
Education.
Ariely, D. (2008). Predictably irrational. New York, NY: HarperAudio.
Asch, S. E. (1951). Effects of group pressure upon the modification and distortion of judgment.
In H. Guetzkow (ed.) Groups, leadership and men. Pittsburgh, PA: Carnegie Press.
Babcock, L., & Loewenstein, G. (1997). Explaining bargaining impasse: The role of self-serving
biases. Journal of Economic Perspectives, 11(1), 109-126.
Baddeley, M. C., Curtis, A., & Wood, R. (2004). An introduction to prior information derived
from probabilistic judgements: Elicitation of knowledge, cognitive bias and herding.
Geological Society, London, Special Publications, 239(1), 15-27.
10.1144/GSL.SP.2004.239.01.02.
HOW COGNITIVE BIASES CAN UNDERMINE SCALE UP
24
Baron, J. (2008). Thinking and deciding. Cambridge: University Press, Cambridge.
Baumeister, R. (2003). The psychology of irrationality: Why people make foolish, self-defeating
choices. The Psychology of Economic Decisions, 1, 3-16.
Berkovich, Izhak. (2011). No we won't! Teachers' resistance to educational reform. Journal of
Educational Administration. 49. 563-578. 10.1108/09578231111159548.
Beshears, J., Choi, J. J., Laibson, D., & Madrian, B. (2006). Early decisions: A regulatory
framework (No. w11920). National Bureau of Economic Research.
Brackett, M. A., Rivers, S. E., Reyes, M. R., & Salovey, P. (2012). Enhancing academic
performance and social and emotional competence with the RULER feeling words
curriculum. Learning and Individual Differences, 22(2), 218-224.
Button, K. S., Ioannidis, J. P., Mokrysz, C., Nosek, B. A., Flint, J., Robinson, E. S., & Munafò,
M. R. (2013). Empirical evidence for low reproducibility indicates low pre-study odds.
Nature Reviews Neuroscience, 14(12), 877-877.
Chambers, D. & Norton, W. (forthcoming). Sustaining Impact after Scaling Using Data and
Continuous Feedback. In List, J., Suskind, D., & Supplee, L. H., The Scale-up Effect in
Early Childhood and Public Policy. Routledge.
Choi, J. J., Laibson, D., Madrian, B.C., & Metrick, A. (2004). For better or for worse: Default
effects and 401(k) savings behavior. In D. A.Wise (Ed.), Perspectives in the economics of
aging, (pp. 81–121). Chicago: University of Chicago Press.
Cho, M. K., & Bero, L. A. (1996). The quality of drug studies published in symposium
proceedings. Annals of Internal Medicine, 124(5), 485-489.
Cohen, J. (1994). The earth is round (p<. 05). American Psychologist, 49, 997-1003.
Dean, M., Kibris, O. & Masatlioglu, Y. (2014). Limited attention and status quo bias. Working
HOW COGNITIVE BIASES CAN UNDERMINE SCALE UP
25
Paper, No. 2014-11, Brown University, Department of Economics, Providence, RI
Davidson, R. A. (1986). Source of funding and outcome of clinical trials. Journal of General
Internal Medicine, 1(3), 155-158.
Dawson, E., Gilovich, T., & Regan, D. T. (2002). Motivated reasoning and performance on the
was on selection task. Personality and Social Psychology Bulletin, 28(10), 1379-1387.
Eidelman, S., & Crandall, C. S. (2014). The intuitive traditionalist: How biases for existence and
longevity promote the status quo. In Advances in experimental social psychology (Vol.
50, pp. 53-104). Academic Press.
Fernandez, R., & Rodrik, D. (1991). Resistance to reform: status quo bias in the presence of
individual-specific uncertainty. The American Economic Review, 1146-1155.
Francis, G. (2013). Replication, statistical consistency, and publication bias. Journal of
Mathematical Psychology, 57(5), 153-169.
Friedberg, M., Saffran, B., Stinson, T. J., Nelson, W., & Bennett, C. L. (1999). Evaluation of
conflict of interest in economic analyses of new drugs used in oncology. Jama, 282(15),
1453-1457.
Gelman, A., & Carlin, J. (2014). Beyond power calculations: Assessing type S (sign) and type M
(magnitude) errors. Perspectives on Psychological Science, 9(6), 641-651.
Gelman, A., & Loken, E. (2014). The statistical crisis in science: data-dependent analysis--a"
garden of forking paths"--explains why many statistically significant comparisons don't
hold up. American Scientist, 102(6), 460-466.
Gigerenzer, G. (1996). On narrow norms and vague heuristics: A reply to Kahneman and
Tversky. Psychological Review, 103(3), 592–596
HOW COGNITIVE BIASES CAN UNDERMINE SCALE UP
26
Gilovich, T. (1983). Biased evaluation and persistence in gambling. Journal of Personality and
Social Psychology, 44(6), 1110-1126. https://doi.org/10.1037/0022-3514.44.6.1110
Gilovich, T., Griffin, D., & Kahneman, D. (2002). Heuristics and biases: The psychology of
intuitive judgment. Cambridge University Press.
Goldsmith, L. A., Blalock, E. N., Bobkova, H., & Hall 3rd, R. P. (2006). Picking your peers. The
Journal of Investigative Dermatology, 126(7), 1429-1430. doi:10.1038/sj.jid.5700387
Halpern, S. D., Loewenstein, G., Volpp, K. G., Cooney, E., Vranas, K., Quill, C. M., ... &
Arnold, R. (2013). Default options in advance directives influence how patients set goals
for end-of-life care. Health Affairs, 32(2), 1-10.
Henshel, R. L. & Johnston, W. (1987). The emergence of bandwagon effects: A theory.
Sociological Quarterly, 28(4), 493-511.
Haselton, M. G., Nettle, D., & Murray, D. R. (2016). The evolution of cognitive bias. In D. M.
Buss (Ed.), The handbook of evolutionary psychology: Integrations (pp. 968-987). John
Wiley & Sons Inc.
Hastorf, A. H., & Cantril, H. (1954). They saw a game; a case study. The Journal of Abnormal
and Social Psychology, 49(1), 129.
Hershfield, H. E. (2011). Future self-continuity: How conceptions of the future self transform
intertemporal choice. Annals of the New York Academy of Sciences, 1235, 30.
https://doi.org/10.1111/j.1749-6632.2011.06201.x
Ioannidis, J. P. (2005). Why most published research findings are false. PLOS Medicine, 2(8),
e124.
Iyengar, S.S. and M. L. Lepper. (2000). When Choice Is Demotivating: Can One Desire Too
Much of a Good Thing? Journal of Personality and Social Psychology, 79(6):995.1006.
HOW COGNITIVE BIASES CAN UNDERMINE SCALE UP
27
Jonas, E., Schulz-Hardt, S., Frey, D., & Thelen, N. (2001). Confirmation bias in sequential
information search after preliminary decisions: an expansion of dissonance theoretical
research on selective exposure to information. Journal of Personality and Social
Psychology, 80(4), 557.
Johnson, E. J., Hershey, J., Meszaros, J., & Kunreuther, H. (1993). Framing, probability
distortions, and insurance decisions. Journal of Risk and Uncertainty, 7(1), 35-51.
https://doi.org/10.1007/BF01065313
Kahan, D. M., Peters, E., Dawson, E. C., & Slovic, P. (2013). Motivated numeracy and
enlightened self-government (Cultural Cognition Project Working Paper No. 116). New
Haven, CT: Yale University Law School.
Kahneman, Daniel and Amor Tversky. (1972). Subjective probability: A judgment of
representativeness. Cognitive Psychology 3(3) pp 430-454.
Kahneman, D. (2011). Thinking, fast and slow. New York, NY: Farrar, Straus and Giroux.
Kahneman, D., Knetsch, J. L., & Thaler, R. H. (1991). Anomalies: The endowment effect, loss
aversion, and status quo bias. Journal of Economic Perspectives, 5 (1): 193-206.
Kahneman, D., & Tversky, A. (1979). Prospect theory: An analysis of decision under risk.
Econometrica, 47(2), 263-292.
Kalil, A., Mayer, S. E., & Gallegos, S. (2020). Using behavioral insights to increase attendance
at subsidized preschool programs: The Show Up to Grow Up intervention. Forthcoming, .
Organizational Behavior and Human Decision Processes.
HOW COGNITIVE BIASES CAN UNDERMINE SCALE UP
28
Kleck, R. E., & Wheaton, J. (1967). Dogmatism and responses to opinion-consistent and
opinion-inconsistent information. Journal of Personality and Social Psychology, 5(2),
249.
Krimsky, S. (2013). Do financial conflicts of interest bias research? An inquiry into the “funding
effect” hypothesis. Science, Technology, & Human Values, 38(4), 566-587.
Kunda, Z. (1990). The case for motivated reasoning. Psychological Bulletin, 108(3), 480.
Lee, C. J., Sugimoto, C. R., Zhang, G., & Cronin, B. (2013). Bias in peer review. Journal of the
American Society for Information Science and Technology, 64(1), 2-17.
Madrian, B. C., & Shea, D. F. (2001). The power of suggestion: Inertia in 401 (k) participation
and savings behavior. The Quarterly Journal of Economics, 116(4), 1149-1187.
Mahoney, M. J. (1977). Publication prejudices: An experimental study of confirmatory bias in
the peer review system. Cognitive Therapy and Research, 1(2), 161-175.
Maniadis, Z., Tufano, F., & List, J. A. (2014). One swallow doesn't make a summer: New
evidence on anchoring effects. American Economic Review, 104(1), 277-290.
Mayer, S. E., Kalil, A., Oreopoulos, P., & Gallegos, S. (2019). Using behavioral insights to
increase parental engagement: The Parents and Children Together Intervention. Journal
of Human Resources, 54(4), 900-925.
McCormack, J., & Greenhalgh, T. (2000). Seeing what you want to see in randomised controlled
trials: versions and perversions of UKPDS data. Bmj, 320(7251), 1720-1723.
Mihalic, S. F., Fagan, A. A., & Argamaso, S. (2008). Implementing the LifeSkills Training drug
prevention program: factors related to implementation fidelity. Implementation Science,
3(1), 5.
HOW COGNITIVE BIASES CAN UNDERMINE SCALE UP
29
Moshinsky, A., & Bar-Hillel, M. (2010). Loss aversion and status quo label bias. Social
Cognition, 28(2), 191-204. https://doi.org/10.1521/soco.2010.28.2.191
Nisbett, R. E. (2015). Mindware: Tools for smart thinking. New York, NY: Farrar, Straus and
Giroux.
Phillips, C. J. (2015). The new math. University of Chicago Press.
Pichert, D., & Katsikopoulos, K. V. (2008). Green defaults: Information presentation and pro-
environmental behaviour. Journal of Environmental Psychology, 28(1), 63-73.
Polites, G. L., & Karahanna, E. (2012). Shackled to the status quo: The inhibiting effects of
incumbent system habit, switching costs, and inertia on new system acceptance. MIS
Quarterly, 21-42.
Quattrone, G. A., & Tversky, A. (1988). Contrasting rational and psychological analyses of
political choice. American Political Science Review, 82(3), 719-736.
Ringwalt, C. L., Ennett, S., Johnson, R., Rohrbach, L. A., Simons-Rudolph, A., Vincus, A., &
Thorne, J. (2003). Factors associated with fidelity to substance use prevention curriculum
guides in the nation's middle schools. Health Education & Behavior, 30(3), 375-391.
Rochon, P. A., Gurwitz, J. H., Simms, R. W., Fortin, P. R., Felson, D. T., Minaker, K. L., &
Chalmers, T. C. (1994). A study of manufacturer-supported trials of nonsteroidal anti-
inflammatory drugs in the treatment of arthritis. Archives of Internal Medicine, 154(2),
157-163.
Rohrbach, L. A., Graham, J. W., & Hansen, W. B. (1993). Diffusion of a school-based substance
abuse prevention program: predictors of program implementation. Preventive Medicine:
An International Journal Devoted to Practice and Theory.
HOW COGNITIVE BIASES CAN UNDERMINE SCALE UP
30
Samuelson, W., & Zeckhauser, R. (1988). Status quo bias in decision making. Journal of Risk
and Uncertainty, 1(1), 7-59.
Sanders, M., Halpern, D., & Service, O. (2013). Applying behavioural insights to charitable
giving.
Simmons, J. P., Nelson, L. D., & Simonsohn, U. (2011). False-positive psychology: Undisclosed
flexibility in data collection and analysis allows presenting anything as significant.
Psychological Science, 22(11), 1359-1366.
Sunstein, C. R. (2014). Why nudge?: The politics of libertarian paternalism. Yale University
Press.
Suri, G., Sheppes, G., Schwartz, C., & Gross, J. J. (2013). Patient inertia and the status quo bias:
when an inferior option is preferred. Psychological Science, 24(9), 1763-1769.
Teichert, T., Ferrera, V. P., & Grinband, J. (2014). Humans optimize decision-making by
delaying decision onset. PlOS One, 9(3).
Terhart, E. (2013). Teacher resistance against school reform: reflecting an inconvenient
truth, School Leadership & Management, 33:5, 486-
500, DOI: 10.1080/13632434.2013.793494
Thaler, R. H. & Sunstein, C.R. (2008). Nudge: Improving decisions about health, wealth, and
happiness. New Haven, CT: Yale University Press.
Trotter, Griffin. (2002) “Why were the benefits of tPA exaggerated?” The Western Journal of
Medicine;176(3):194-197.
Tversky, A., & Kahneman, D. (1974). Judgment under uncertainty: Heuristics and biases.
Science, 185(4157), 1124-1131.
HOW COGNITIVE BIASES CAN UNDERMINE SCALE UP
31
Vaganay, A. (2016). Outcome reporting bias in government-sponsored policy evaluations: A
qualitative content analysis of 13 studies. PLOS One, 11(9).
https://doi.org/10.1371/journal.pone.0163702
Van der Meer, T. W., Hakhverdian, A., & Aaldering, L. (2016). Off the fence, onto the
bandwagon? a large-scale survey experiment on effect of real-life poll outcomes on
subsequent vote intentions. International Journal of Public Opinion Research, 28(1), 46-
72.
Wason, P. C. (1960). On the failure to eliminate hypotheses in a conceptual task. Quarterly
Journal of Experimental Psychology, 12(3), 129-140.
Wason, P. C. (1968). Reasoning about a Rule. Quarterly Journal of Experimental Psychology,
20(3), 273-281.
World Bank. (2015). World development report 2015: Mind, Society, and Behavior. World
Bank.
York, B. N., & Loeb, S. (2014). One step at a time: The effects of an early literacy text
messaging program for parents of preschoolers (No. w20659). National Bureau of
Economic Research.
Zuckerman, M. (1979). Attribution of success and failure revisited, or: The motivational bias is
alive and well in attribution theory. Journal of Personality, 47(2), 245-287.