Intervention Tournaments: An Overview of Concept,Design,and Implementation

Abstract

A large portion of research in the social sciences is devoted to using interventions to combat societal and social problems, such as prejudice, discrimination, and intergroup conflict. However, these interventions are often developed using the theories and/or intuitions of the individuals who developed them and evaluated in isolation without comparing their efficacy with other interventions. Here, we make the case for an experimental design that addresses such issues: an intervention tournament—that is, a study that compares several different interventions against a single control and uses the same standardized outcome measures during assessment and participants drawn from the same population. We begin by highlighting the utility of intervention tournaments as an approach that complements other, more commonly used approaches to addressing societal issues. We then describe various approaches to intervention tournaments, which include crowdsourced, curated, and in-house-developed intervention tournaments, and their unique characteristics. Finally, we discuss practical recommendations and key design insights for conducting such research, given the existing literature. These include considerations of intervention-tournament deployment, characteristics of included interventions, statistical analysis and reporting, study design, longitudinal and underlying psychological mechanism assessment, and theoretical ramifications.

Complicated problems, such as violent conflicts, inequality, mass migration, climate change, vaccine hesitancy, and global pandemics, necessitate extraordinary efforts to resolve. As a prominent example, the novel coronavirus, or COVID-19, has affected the lives of almost all human beings, infecting and killing millions (World Health Organization [WHO], n.d.) and severely damaging the world economy (Ayittey et al., 2020). COVID-19 was found to be highly contagious and transmitted from asymptomatic infected individuals (Nishiura et al., 2020; Rothe et al., 2020). Thus, epidemiologists argued that curbing the pandemic would be challenging without the development of an effective vaccine (Chen et al., 2020; Sah et al., 2021), which led to a race to develop one; the first vaccines were approved in December 2020 (e.g., Baden et al., 2021; Callaway, 2020; Polack et al., 2020). This race included dozens of labs around the world that developed and tested potential vaccines in parallel by using their own theories, outcome measures, and benchmarks and focusing on whether their approach was effective and why (Callaway, 2020; Le et al., 2020). At the same time, given the urgency of the situation, many researchers and policymakers pushed for a collaborative approach. Correspondingly, on April 9, 2020, WHO (2020) published a call for a single, large, and international vaccine “tournament” of the vaccines developed at different labs to identify the most effective vaccine compared with a single placebo condition. This call is a useful example of the alternative, result-oriented approach that has gained prominence in the study of medicine to address pressing, complicated, and costly health-related problems (see Adaptive Platform Trials Coalition, 2019; Parmar et al., 2014; Wason et al., 2016), although, in the case of the COVID-19 vaccine, it was eventually not endorsed by the research community. We argue that in the social sciences, however, this approach has been underused.

Thus, the aim of the current article is to make the case for and describe the utility of an “intervention tournament” in the psychological sciences to address societal problems. Intervention tournaments (which are sometimes also called “multiarm trials,” “intervention contests,” “comparative evaluations,” or “megastudies”) are studies that compare several different interventions against a single control condition with the same standardized outcome measure(s) and participants drawn from the same population (Bruneau et al., 2018; Efentakis & Politis, 2006; Lai et al., 2014; Milkman et al., 2021; Parmar et al., 2014; for elaborate extensions of this approach used in medical research, see Adaptive Platform Trials Coalition, 2019; Wason et al., 2016). We begin by identifying the two main approaches used to advance solutions and, in particular, interventions to pressing social and medical problems. Then, we focus on intervention tournaments by providing its definition and main types. This is followed by a nuanced discussion that includes some practical considerations and recommendations and the benefits and limitations for conducting psychological intervention tournaments in the lab and field settings to address societal problems.

Two Approaches to Intervention Development

The above example of how researchers, clinicians, practitioners, and policymakers addressed the COVID-19 vaccine development suggests that there are two different approaches in which pressing problems can be addressed, which we refer to as “top-down” and “bottom-up” for sake of simplicity. In contrast to the bottom-up approach that can be done with intervention tournaments, on which we elaborate in the following, the top-down approach—sometimes called the “mechanism-in-isolation design” (Lai et al., 2014) or “parallel-group randomized control trials” (Adaptive Platform Trials Coalition, 2019)—is the more standard, wildly used, “status quo” approach and has generally been considered the “gold standard” approach (e.g., Adaptive Platform Trials Coalition, 2019; Concato et al., 2000; Kratochwill & Levin, 2014; Slade & Priebe, 2001; Wilson & Juarez, 2015). It entails the development and assessment of an intervention that is grounded in a specific theoretical approach, and its goal is to assess if a specific intervention works and why.

The development of this approach generally progresses in the following manner. First, researchers develop a theory to explain a phenomenon. For example, research that is based on the current coronavirus, as well as on previous SARS and MERS outbreaks, found that one of the main reasons why the coronavirus is highly contagious is because of its spikes (which also gives the virus its name, corona, Latin for “crown”), which can attach to particular proteins in human airway cells (Li, 2016; Tian et al., 2020). Second, following this molecular understanding of the disease, researchers and practitioners then developed an intervention according to their theoretical understanding of coronavirus biology. For example, given the knowledge of the coronavirus spikes, some labs tried to develop vaccines aimed at exposing the body to a spike protein, which would cause the immune system to recognize it as an antigen, or foreign entity (Callaway, 2020; Le et al., 2020; Liu et al., 2020). Third, researchers and practitioners tested this intervention in isolation compared with a control (placebo) condition and in some cases to a second intervention that was previously found to be effective. For example, when developing a vaccine, there is an agreed protocol (with some variants; see e.g., Adaptive Platform Trials Coalition, 2019; Bothwell et al., 2016) that needs to be followed closely to prove a vaccine is effective and safe for use by the general public. This approach includes random assignment between the treatment and placebo groups.

We would argue that although the process of developing most psychological interventions is not identical, it is similar. For example, recent research conducted by Bruneau and colleagues (Kteily et al., 2016; Moore-Berg, Ankori-Karlinsky, et al., 2020; Moore-Berg, Hameiri, & Bruneau, 2020; see also Lees & Cikara, 2020, 2021; Ruggeri et al., 2021) identified that overly pessimistic metaperceptions—or how one thinks out-group members view one’s in-group—are prominent psychological factors that feed intergroup hostility. Specifically, individuals tend to think that adversarial out-group members hold much more negative views of the in-group than they do in reality. For example, Moore-Berg, Ankori-Karlinsky, et al. (2020) found that Democrats and Republicans in the United States think that people from the out-group party dislike and dehumanize them at least twice as much as they actually do, which is strongly associated with antipathy and spiteful policy support that comes at the expense of the country. Given this theoretical understanding, several different interventions that aim to correct overly pessimistic metaperceptions were developed and assessed vis-à-vis a control group (Kteily et al., 2016; Lees & Cikara, 2020; Mernyk et al., 2022). For example, in one study in the context of partisan polarization in the United States, Lees and Cikara (2020) developed an intervention in which they showed participants the true value of the partisan out-group perceptions together with the participants’ perceived metaperceptions. Compared with a control condition in which participants were just reminded of their own metaperceptions, the intervention successfully reduced negative metaperceptions and, consequently, negative motivational attributions toward the out-group. This original study was then replicated in nine additional countries, which established the robustness of this psychological intervention (Ruggeri et al., 2021).

Although the benefits of this top-down approach should not be discarded, we argue that it does entail several drawbacks. First, when such studies are conducted, the ability to compare interventions across different studies is challenging. That is, it is unclear whether different studies that compare target interventions with different control groups are comparable, which makes it challenging to determine whether a given intervention is more (or less) effective compared with other interventions. Second, top-down mechanism-in-isolation studies do not necessarily rely on standardized outcome measures and often are examined at different times with different outcome measures. Thus, it is difficult to determine what the most relevant outcomes are and whether these same outcomes would be equally affected by other interventions that aim to ultimately achieve the same goal. These outcomes are based on assumptions and decisions made by the researchers themselves that in some cases could be contested and challenged (Paluck et al., 2019; Pettigrew & Hewstone, 2017). This is exacerbated by what Pettigrew and Hewstone (2017) termed as the “single factor fallacy,” which is the tendency for researchers to rely on their own work, based on their specific theoretical framework and variables they focus on, to develop and assess interventions. As a consequence, important factors, including alternative explanations and theories, or other important variables, may be overlooked. Finally, and as mentioned above, the top-down approach is rather limited in its ability to tackle complicated problems quickly and efficiently because it is based on the available resources each lab has and each lab compares the intervention or interventions with different and unstandardized control groups (e.g., Adaptive Platform Trials Coalition, 2019; Bothwell et al., 2016; Wilson & Juarez, 2015).

Conversely, the bottom-up approach is result-oriented and focuses on finding a solution to a problem by identifying what is effective in the most cost-effective and efficient manner possible with a set of agreed-on outcome measures. This can be done with intervention tournaments. An intervention tournament compares several different interventions against the same control and standardized outcome measures. Interventions can be selected using different criteria (e.g., established theory), as we elaborate on below. The interventions are then assessed with participants drawn from the same population to isolate which intervention or interventions is most effective. The main goal of an intervention tournament is to identify the most successful approach or approaches for mitigating the problem at hand, which means that establishing the mechanism through which one intervention is effective is secondary to the main goal of this approach. Although this approach has gained prominence in medical research, as the WHO (2020) call for a single COVID-19 vaccine tournament exemplifies, to the best of our knowledge and as mentioned above, in the social sciences, this approach is still relatively underused (but see Axelrod & Hamilton, 1981; Bruneau et al., 2018; Lai et al., 2014, 2016; Milkman et al., 2021). Thus, with the aim of facilitating the use of intervention tournaments in the social sciences, in the heart of the current article, we define this approach, review common features and issues when intervention tournaments are used, identify different types of intervention tournaments, and offer recommendations to promote best practices and avoid potential pitfalls in their design, deployment, and reporting.

Intervention Tournaments

Broadly speaking, an intervention tournament is an experiment that tests and evaluates the causal effects of different approaches, or interventions, on a set of outcome measures with participants drawn from the same population (Bruneau et al., 2018; Lai et al., 2014, 2016; Milkman et al., 2021; Parmar et al., 2014). As mentioned above, the goal of an intervention tournament is to assess what is effective in addressing a social problem compared with a single control condition. Therefore, the question of why an intervention is effective is secondary. However, given its importance, the mechanism or mechanisms can be preliminarily examined as part of the intervention tournament and then more thoroughly tested with follow-up studies, as we elaborate on below. In other words, the goal of intervention tournaments is to select interventions that researchers think might work to address pressing problems (see elaboration on inclusion criteria later in the article) and test their initial effectiveness. Then, if effective, researchers can work backward to identify why they worked and what the boundary conditions of the successful intervention or interventions are.

We argue that there are several benefits to the intervention-tournament approach. First, it is efficient because it allows for the comparison of numerous interventions in a fast and cost-effective manner to identify the most effective intervention or interventions, if any (Freidlin et al., 2008). Second, it uses a standardized approach within a given tournament to assess the interventions such that the interventions are measured with the exact same outcome measures and with participants from the same population. As elaborated on below, we suggest that there can be considerable variations between different intervention tournaments. Third, when conducted among a large and diverse sample, intervention tournaments can also identify potential moderators, which shows that different interventions are effective for individuals with different characteristics, which can be a stepping-stone in designing personalized interventions (e.g., Bar-Tal & Hameiri, 2020; Bruneau, 2015; Collins & Varmus, 2015; Cuijpers et al., 2016; Halperin & Schori-Eyal, 2020; Hirsh et al., 2012).

In the next section, we describe three distinct types of intervention tournaments that are based on either crowdsourcing (e.g., Axelrod & Hamilton, 1981; Bennett & Lanning, 2007; Forscher et al., 2020; Lai et al., 2014, 2016; Milkman et al., 2021; Uhlmann et al., 2019; see also WHO, 2020), curating (e.g., Bruneau et al., 2018; Moore-Berg, Hameiri, Falk, & Bruneau, 2022), or in-house development of interventions (e.g., Bruneau et al., 2022; Van Assche et al., 2020; see also Efentakis & Politis, 2006).

Crowdsourced intervention tournaments are perhaps the most promising type of intervention tournaments and are also in line with recent calls for more collaborative science (e.g., Forscher et al., 2020; IJzerman et al., 2020; Moore-Berg, Bernstein, et al., 2022; Uhlmann et al., 2019). A crowdsourced intervention tournament calls on scientists, practitioners, media experts, and so on to submit intervention ideas to be assessed within a single context (for a similar approach, the Metaketa Initiative, that examines the same intervention compared with numerous control groups and in some cases alternative interventions across multiple diverse geographic regions, see Leaver, 2019). For example, to promote flu (and potentially COVID-19) vaccinations in the United States, Milkman et al. (2021) crowdsourced 19 different nudge interventions created by 26 behavioral scientists. They tested these different nudges against one control group on one standardized outcome (i.e., receiving the flu shot) in an intervention tournament. They found that out of the 19 interventions, six significantly increased the percentage of participants who received the flu shot by approximately 3% to 4% compared with the control (42%) and that on average, all 19 nudges increased vaccination levels by 2.1%.

As a second example, Lai et al. (2014, 2016) held a series of crowdsourced intervention tournaments to identify the interventions that are the most effective at reducing implicit racial biases. To do this, Lai and colleagues sent out a call for different labs to propose what they believe to be the most effective intervention to reduce implicit racial bias. They then received 17 interventions in total that were based on a diverse set of hypothesized mechanisms (e.g., exposure to positive exemplars, imagined contact, perspective taking, inducing empathy). Lai and colleagues compared all of these interventions with a no-intervention control condition in an iterative process—researchers who contributed interventions to the tournament were able to modify their intervention on the basis of the results of previous rounds of the tournament (i.e., studies), which then led to greater effectiveness in reducing implicit intergroup bias across the studies (for a similar approach, see Axelrod, 1980a, 1980b; Axelrod & Hamilton, 1981). Across these four intervention-tournament studies, Lai et al. (2014) found eight interventions to be most effective at reducing implicit racial bias. In a follow-up intervention tournament, Lai et al. (2016) found that all eight successful interventions from Lai et al. (2014) were again effective in reducing implicit racial bias when assessed immediately after the interventions but that these effects did not persist when the participants were reassessed several hours to days later. We return to the notion of replicating original results and the longitudinal aspect of intervention tournaments in the next section.

Note that these crowdsourced intervention tournaments are not limited to researchers and research labs, which will likely submit interventions on the basis of their own work and theoretical framework (see Pettigrew & Hewstone, 2017). Crowdsourced intervention tournaments can also be used to solicit interventions developed by practitioners, media experts, filmmakers, and so on. These individuals can develop potentially effective and engaging interventions that rely on their creativity, expertise, experience, intuition, and contextual knowledge (Bar-Tal & Hameiri, 2020; Bruneau et al., 2022; IJzerman et al., 2020). Furthermore, crowdsourcing interventions from those outside of academia can improve the external validity of the research by incorporating interventions already used in the field.

Curated intervention tournaments have gained increased attention in recent years through the work of Bruneau and colleagues (2018; see also Moore-Berg, Hameiri, Falk, & Bruneau, 2022). In this approach, the researchers themselves curate different interventions that they believe have the potential to mitigate a specific problem. Specifically, the curators can take real-world interventions that others, mostly practitioners, have been using, test them in an intervention tournament, and then identify why they were effective, if indeed they were. For example, in Bruneau and colleagues’ work, they included various interviews, documentary segments, and news clips available in mainstream media that they—and the practitioners they consulted—thought would reduce levels of Islamophobia. Thus, the interventions themselves are generally created and disseminated before assessment, whether it is in the mainstream media, social media, or elsewhere, and are developed on the basis of the creators’ intuition of what would constitute an effective intervention. However, although these creators may develop compelling and engaging content, they rarely rigorously test the effectiveness of these materials in achieving the desired outcomes (Davidson, 2017).

For example, Bruneau et al. (2018) conducted an intervention tournament to identify videos that most effectively reduce the tendency of non-Muslims to collectively blame all Muslims for the actions of individual Muslim extremists. The underlying theoretical assumption is that collective blame increases Islamophobia among non-Muslims by feeding negative and aggressive attitudes and behaviors toward Muslims. Therefore, the aim of this intervention tournament was to find the most effective intervention to short-circuit this tendency. Thus, Bruneau at al. curated videos that were created by both Muslim and non-Muslim practitioners. The researchers chose these videos because they were both compelling and diverse in terms of styles of delivery (didactic, narrative, satire) and their underlying theories (that were mapped onto them a priori by the researchers). Bruneau et al. found that a 2-min video that showed an interview with a Muslim American woman discussing the tendency of non-Muslim Americans to blame all Muslims for terror attacks but not blame Christians for extremism by individual Christians was most effective at reducing collective blame among non-Muslims and, correspondently, hostility toward Muslims. Bruneau and colleagues then replicated these results in several follow-up studies using a conceptually similar intervention (Bruneau et al., 2018, 2020).

In a supplemental study, Bruneau et al. (2018) asked an independent set of participants to rate the extent to which they thought the videos would be most effective at reducing Islamophobia. Participants underestimated the potential effect of this particular collective blame video and forecasted that it would be much less effective in reducing collective blame of Muslims compared with the obtained effects. This highlights the need to rigorously test the effectiveness of interventions rather than rely solely on the intuition of researchers, laypeople, and the people who create these interventions and materials. Many of the curated interventions Bruneau and colleagues used in their research (see also Gallardo et al., 2022; Moore-Berg, Hameiri, Falk, & Bruneau, 2022) were developed by practitioners to promote their goal to reduce Islamophobia in the United States but were never experimentally tested. The results of these curated intervention tournaments highlight the importance of conducting this rigorous testing.

Finally, in-house-developed intervention tournaments tend to have a somewhat different goal than crowdsourcing and curated intervention tournaments. In this approach, the interventions created for the intervention tournament are in many cases based on a single theory because they are the product of one group of researchers or a single lab. The different interventions in the tournament can be, for example, a series of interventions with different underlying mechanisms but the same goal, or they can be different iterations of the same intervention with the same underlying goal. As an example from outside of the social sciences, researchers might design different iterations of polymer networks of the same drug to determine the most effective drug-delivery system (Efentakis & Politis, 2006). In the social sciences, this can be, for example, manipulating the delivery techniques and media style (e.g., providing guided self-help via face-to-face meetings vs. via email to address eating disorders; Jenkins et al., 2021) of the same source materials or examining the optimal dose of an intervention (e.g., one vs. three sessions of exposure therapy to prevent the development of posttraumatic stress disorder; Maples-Keller et al., 2020).

For example, Bruneau et al. (2022) created a series of 10 different video interventions aimed at promoting more conciliatory views of FARC ex-combatants among non-FARC Colombians. ¹ All of the videos were created by the researchers in collaboration with local filmmakers and were based on the same source material, which included interviews with FARC ex-combatants and their non-FARC Colombian neighbors in a rural demobilization camp. The interviews addressed non-FARC Colombians’ misperceptions of FARC ex-combatants’, including their willingness or unwillingness to let go of violence and reintegrate into mainstream Colombian society. Thus, the core of all the videos highlighted evidence of the successful coexistence between these demobilized FARC members and their local neighbors. The main variation between the videos was the interviewee (i.e., ex-combatants, non-FARC Colombians, or both) and the order of their appearance (i.e., ex-combatants before non-FARC Colombians or vice versa). Bruneau et al. found that most videos were effective in reducing anti-FARC beliefs and attitudes among non-FARC Colombians and increased support for pro-FARC policies and for the peace process between non-FARC and ex-FARC Colombians. However, one video, which focused on FARC ex-combatants’ unwillingness and ability to change and reintegrate into the Colombian society and included both FARC ex-combatants and non-FARC responses (in this order), was the most effective both immediately and longitudinally approximately 10 to 12 weeks later. In the intervention tournament and in preregistered follow-up studies, in which these results were replicated, Bruneau et al. also provided some insight into the underlying psychological mechanism. It seemed that this video was effective because it changed participants’ belief about FARC ex-combatants’ ability to change (Goldenberg et al., 2018; Halperin et al., 2011), but not affect toward them, and showed an effect on a behavioral measure that can promote peace and reintegration of FARC ex-combatants.

This highlights the (in-house-developed) intervention tournament as a tool to test different iterations of a similar intervention to zero in on the most effective one. In other words, in most cases, the core is similar, whether it is a drug (Efentakis & Politis, 2006) or the psychological content (Bruneau et al., 2022; Jenkins et al., 2021; Maples-Keller et al., 2020; Milkman et al., 2011; Rosler et al., 2021) that is administered, but the delivery is different in each of these interventions.

Note that in-house-developed intervention tournaments are not limited to testing different iterations of the same underlying intervention and can also include interventions that are based on completely different underlying mechanisms, which are tested against each other (e.g., Hameiri et al., 2018; Van Assche et al., 2020; Yokum et al., 2018) and sometimes against their combination (Kim et al., 2021; Moore-Berg, Hameiri, & Bruneau, 2022; Rosler et al., 2021). For example, Yokum et al. (2018) examined the efficacy of different variations of letters that remind Medicare recipients to get the flu vaccine. These letters were based on different psychological mechanisms and included messages with implementation-intention prompts and enhanced active choice. Yokum et al. found that all letters, regardless of their underlying psychological approach, increased vaccination rates compared with the control condition.

Key Design Insights From Past Intervention-Tournament Research

Regardless of which intervention tournament type is chosen, there are various design considerations that researchers should take into account when they develop their intervention tournament. These considerations are aggregated across the existing intervention-tournament literature and provide important design insights into what makes intervention tournaments successful.

Intervention-tournament viability

Researchers who have used intervention tournaments have taken several approaches to intervention curation and inclusion, as reviewed above, which include crowdsourcing, curating from available materials, or developing interventions in-house. However, before intervention procurement and tournament deployment, the first question that researchers need to address is whether it is at all suitable to conduct an intervention tournament to address the research problem at hand. Indeed, the suitability of an intervention tournament depends on the type of tournament in question. In cases of crowdsourced and curated (and to a lesser extent, in-house-developed) intervention tournaments, the main criterion for conducting one is the wealth of existing theories and practices and a collective urgency to address a particular societal problem. In other words, for an intervention tournament to be viable, it needs to address a problem that many feel is important and contain interventions based on prior research (in case of researchers) or developed materials that were (most likely) not empirically examined before (in case of practitioners). It should come as no surprise, then, that all of the examples we provide throughout this article include pressing global challenges such as prejudice, intergroup conflicts, polarization, Islamophobia, vaccine hesitancy, and so on.

We argue that another criterion for a viable intervention tournament is that the problem that is being addressed is challenging and includes overcoming different psychological barriers (e.g., Bar-Tal & Hameiri, 2020; Hornsey & Fielding, 2017). Challenging problems necessitate diverse contributions (which can be crowdsourced or curated) and out-of-the-box thinking that can address the problem at hand from multiple angles (Uhlmann et al., 2019; Van Bavel et al., 2020). Such problems can also increase the motivation of researchers and practitioners to prove that they can come up with the most efficient solution (e.g., Axelrod & Hamilton, 1981; Bennett & Lanning, 2007; Lai et al., 2014, 2016), which can then be put to the test in a crowdsourced intervention tournament. Finally, addressing challenging problems can also benefit from in-house-developed intervention tournaments, especially when there is an approach that shows promise but needs more fine-tuning to, for example, better circumvent psychological barriers and resistance (e.g., Bruneau et al., 2022).

Soliciting and incentivizing intervention submissions

One important issue specifically relevant to crowdsourced intervention tournaments is the process of soliciting and incentivizing intervention submissions. Although, as mentioned above, conducting an intervention tournament that addresses a pressing and challenging societal problem can increase the motivation of researchers and practitioners to take part in an intervention tournament, there are several reasons why people might still be reluctant to submit interventions to crowdsourced tournaments. These mostly include limited time and insufficient motivation and incentives to take part in big-team science (see Forscher et al., 2020; Uhlmann et al., 2019).

Organizers of crowdsourced intervention tournaments have used different approaches to address these potential problems. Some have provided the deserved recognition of the winner or winners of and other participants in the crowdsourced tournament, such as Anatol Rapaport, who won a tournament that aimed to identify the most effective approach in an iterated Prisoner’s Dilemma game (Axelrod, 1980a, 1980b; Axelrod & Hamilton, 1981; granted, providing due credit and recognition is also important in curated intervention tournaments). In other cases, intervention-tournament organizers award prizes (e.g., monetary, authorship) to intervention developers that withstand the selection process and are admitted into the tournament (see Strengthening Democracy Challenge, 2021) or only to the winners and runners-up (e.g., the Netflix Prize; Bennett & Lanning, 2007). In many cases, tournament organizers offer participants authorship on the main publication that results from the intervention tournament (e.g., Lai et al., 2014, 2016; Parmar et al., 2014). Indeed, some provide recognition, prizes, and authorship on academic publications as incentives (see Strengthening Democracy Challenge, 2021).

Moreover, hypothetically, as an additional incentive, it is also possible to provide participants with the opportunity to write a stand-alone article that pertains to their proposed intervention using the intervention-tournament data (see e.g., Leaver, 2019; Parmar et al., 2014). However, in many cases, this is not feasible because the main intervention-tournament article normally includes all interventions and associated data, which means that for such a stand-alone article to be warranted, researchers will need to develop their own novel research questions and hypotheses (e.g., examining moderations that were not previously investigated). Another associated limitation of the intervention-tournament approach is that this common incentive scheme can in fact decrease potential participants’ (in particular, researcher participants) motivation to submit what they might perceive as novel interventions, given that the intervention tournament might decrease their chances of publishing a separate publication in a top-tier journal because it will no longer be novel after being published in the intervention-tournament article.

Finally, although a thorough discussion on potential issues that pertain to authorship are beyond the scope of the current article, we find that in many cases, the intervention-tournament organizers are listed as either first or last authors and all other contributors are listed in an a priori agreed-on order (e.g., in order of effectiveness; see, e.g., Contest Study for Reducing Discrimination in Social Judgement, 2021) or alphabetically in between (e.g., Lai et al., 2014, 2016). In agreement with recent calls for collaborative science, it is advisable that to provide the appropriate intellectual credit that is due to each contributor, a thorough and thoughtful contribution statement should be added using, for example, the CRediT taxonomy (McNutt et al., 2018; for a thorough discussion, see Forscher et al., 2020; Uhlmann et al., 2019).

Inclusion criteria and amount of interventions

One question that might arise during intervention curation and deployment is how many interventions to include and what the inclusion criteria should be (for a relevant discussion in medical research, see Lee et al., 2019; Stallard et al., 2009). Deciding what to include in an intervention tournament is not an easy task and is influenced by various factors that are not necessarily under the control of the researchers, such as available resources and number of submissions in crowdsourced tournaments. Which interventions to include could be decided on the basis of the expertise, intuition, and knowledge of the field of the researchers themselves (e.g., Bruneau et al., 2022); by consulting practitioners (e.g., Gallardo et al., 2022); or, more rigorously, through a peer-review process (Strengthening Democracy Challenge, 2021).

The most important inclusion criterion is to include only interventions that have an underlying theoretical basis that differentiates one from the other (whether it is in the content or delivery mode) and warrants their inclusion in the tournament (e.g., Bruneau et al., 2018; Gallardo et al., 2022; Lai et al., 2014, 2016; Moore-Berg, Hameiri, Falk, & Bruneau, 2022). Practically, in case of multiple submissions of a similar intervention, crowdsourced intervention-tournament organizers can, for example, ask submitters to develop the intervention collaboratively or select one submitter according to different criteria, such as having previous publications, publications on the intervention, or initial compelling data that the intervention is successful (see Strengthening Democracy Challenge, 2021).

A second important selection criterion is whether the proposed intervention is expected to affect the tournament’s main outcome variable or variables. This expected efficiency can be determined on the basis of previous literature on the intervention, and there is a special emphasis on studies (in most cases, smaller scale studies) that were conducted preferably in the same context, using the same population, and with the same (or comparable) outcome measures (cf. Lee et al., 2019). For example, if a researcher wants to reduce interpartisan animosity in the United States, a strong intervention contender could be Lees and Cikara’s (2020) metaperception-correction intervention described above. This is because Lees and Cikara found their intervention to be effective at reducing a related outcome measure (support for purposeful obstructionism among partisans) in the same context among a similar population—an effect that was later replicated in many contexts across the globe (Ruggeri et al., 2021).

A third criterion, which is also somewhat at odds with the previous one, is the degree of novelty of the intervention. This novelty is vis-à-vis other competing interventions in the intervention tournament or the current state of the art in research and in practice. As mentioned above, intervention tournaments are an efficient way to examine the effectiveness of interventions (e.g., Freidlin et al., 2008). Thus, in some cases, such as when testing interventions in a context that has not been heavily researched (e.g., using a curated intervention tournament) or trying to refine an already promising intervention (e.g., using an in-house-developed intervention tournament), intervention tournaments allow researchers and practitioners to experiment with novel ideas.

In this case, intervention tournaments can test different ideas that have only an intuitive appeal or anecdotal evidence to support their effectiveness that can be developed specifically for the tournament or curated for it (e.g., Bruneau et al., 2018). It can derive interventions that were established and shown to be effective in one context and population with a set of outcome measures (e.g., self-affirmation intervention to increase group-based guilt, tested in the context of the Israeli–Palestinian conflict and Bosnia and Herzegovina; Cˇehajicć-Clancy et al., 2011) and test them in another context and population with a related set of outcome measures (e.g., improving intergroup attitudes following the Paris and Brussels terror attacks; Van Assche et al., 2020). Finally, novelty can be injected by implementing different principles from other literatures (e.g., on attitude change and persuasion to circumvent resistance to conflict resolution) to test different iterations of an already promising intervention to find the most effective iteration of it (Bruneau et al., 2022; see also Efentakis & Politis, 2006).

We argue that although it makes sense to include as many interventions as possible in the tournament, the number of tested interventions should be based on the resources available to run an intervention tournament that is sufficiently statistically powered to detect differences between each of the different interventions and the control condition (see elaboration below). In practice, intervention tournaments typically vary in the number of interventions included. For instance, whereas Lai et al. (2014) included 18 crowdsourced interventions in their initial tournament (including a sham intervention), Van Assche et al. (2020) included three in-house-developed interventions.

Potential issues with intervention-inclusion criteria

Intervention tournaments can sometimes include interventions that are diverse in terms of the underlying psychological content, modes of delivery, length, level of engagement, and so on. In some cases, these differences are inevitable because they are an inherent part of the intervention. For example, when it comes to reducing implicit prejudice (Lai et al., 2014, 2016), some researchers might argue that the best way to address this issue is by teaching individuals a new skill that can help them respond empathically, which addresses an important aspect of prejudice. Other researchers might argue that to effectively combat prejudice, one has to provide inconsistent information to change people’s attitudes regarding the prejudiced group (see Bar-Tal & Hameiri, 2020; Hameiri et al., 2014). This can be done, for example, by facilitating some form of intergroup contact (e.g., in person, vicarious, or imagined; e.g., Crisp & Turner, 2012; Dovidio et al., 2017; Pettigrew & Tropp, 2006). These types of interventions will undoubtedly be operationalized differently. In one, participants might be asked to participate in a 15-min-long session in which they learn and exercise a new tool that can help them express more empathy (see the work on cognitive reappraisal as an acquired tool to promote better intergroup relations; e.g., Halperin et al., 2014; Hurtado-Parrado et al., 2019). In the other, they might be asked to passively watch—or actively watch when it comes to virtual reality—a short 2- to 5-min video that includes members of the prejudiced out-group (e.g., Bruneau et al., 2018; Hasson et al., 2019).

Granted, these differences are sometimes unavoidable and in fact might provide critical information about the effectiveness of the different interventions as real-world interventions (see e.g., Bar-Tal & Hameiri, 2020). For example, a lengthy intervention might be less successful, which would indicate that although the psychological content has the potential to reduce prejudice, people lose interest or do not have the motivation to go through a longer intervention, which ultimately renders the intervention to be less effective (e.g., Tamir, 2009; Tamir et al., 2020). In other cases, in which the interventions are curated or created in-house, it is likely that the characteristics (e.g., mode of delivery, length) of the interventions will be easier to control. In these cases, it is advisable to include interventions with similar characteristics to reduce a potential confound. However, these characteristics can also be controlled in a crowdsourced intervention tournament (see e.g., Lai et al., 2014; Strengthening Democracy Challenge, 2021). For example, guidelines can request that all submitted interventions be ethical (e.g., that they do not include any deception), be completed in less than a specific amount of time (e.g., 5 min), be online, and be costless (e.g., that they do not provide any additional monetary incentives for participation).

Although this approach has its merits, because it allows for an intervention tournament that creates an even playing field for the participating interventions (and reduces the risk of potential confounds, as mentioned above), it also points to a limitation in the intervention-tournament approach that should be noted. Although some variation across interventions might be acceptable, most intervention tournaments compare short (or light touch) interventions, which may have the potential to be scaled up relatively easily but might also be less effective, especially across time, than longer and more intense interventions (Paluck et al., 2021), such as months-long in-person or virtual-contact interventions (e.g., Bruneau, Hameiri, et al., 2021; Mousa, 2020). Contact interventions can still be included in an intervention tournament; however, for them to be comparable with other interventions, they are normally parasocial or vicarious (as opposed to in person) contact interventions (e.g., Gallardo et al., 2022). Although longer, more intense interventions can be tested in intervention tournaments, it is much rarer and usually done as part of an in-house-developed intervention tournament and in collaboration with organizations that provide the infrastructure for such a complicated endeavor either in educational, organizational, or clinical settings, in most cases (Jenkins et al., 2021; Leventhal et al., 2015; Maples-Keller et al., 2020). As an example, Leventhal et al. (2015) examined different months-long curricula—including two separate curricula and their combination compared with an active control—to promote resilience and well-being among girls in 76 schools in Bihar, India.

Deploying intervention tournaments

Following intervention procurement, the researchers are then tasked with considering how to deploy the intervention tournament. In most cases, a lab setting might be more feasible in terms of available resources. It can also be more useful because it allows researchers to have more control over the experimental design (Wilson et al., 2010), but potentially at the expense of external validity (e.g., Eastwick et al., 2013; Mitchell, 2012). This includes assuring that there will be no spillover between the conditions, which is more likely to occur when the study includes numerous conditions and all participants are sampled from the same population. Intervention tournaments can also be conducted in field settings. For instance, researchers could partner with organizations that deploy interventions in the community and work with them to develop an intervention tournament with a randomized control design (see e.g., Milkman et al., 2021; Yokum et al., 2018). On top of the clear benefits to external validity, this type of deployment can foster collaborations across both laboratories and practitioners, which creates easier access to testing the intervention in hard-to-reach places (see Acar et al., 2020; Bar-Tal & Hameiri, 2020; Moore-Berg, Bernstein, et al., 2022). However, field intervention tournaments should be conducted with extra care to avoid doing more harm than good, especially when the interventions that are tested do not have a clear track record that establishes their effectiveness (e.g., in curated intervention tournaments).

Comparing intervention efficiency

Following the curation and testing of the interventions, researchers are then tasked with deciding how to compare the interventions. There are three important decisions to make: (a) what the outcome measure or measures are, (b) what statistics to compare, and (c) what type of control condition or conditions to include in the intervention tournament. As mentioned above, one of the first decisions the researchers make when considering whether to deploy an intervention tournament is what problem (ranging from concrete to abstract) to address. This decision then directly translates to the specific outcome measures that are included in the tournament. Concrete problems are usually operationalized as one specific outcome measure. For example, Milkman et al. (2021) attempted to increase flu vaccination by using nudging text messages. Therefore, their only outcome measure was the extent to which their participants got vaccinated. Likewise, Lai et al. (2014) attempted to reduce implicit prejudice and therefore focused on participants’ Implicit Association Test (IAT) scores as an outcome measure (although they did include one additional measure of self-reported racial attitudes, the tournament’s winners were decided using only the IAT scores).

Abstract problems, on the other hand, usually mean that different outcome variables are measured that reflect different operationalizations of the abstract problem. These, in many cases, include behavioral and policy-relevant measures (in addition to attitudinal or affective measures) that are more closely related to the problem. For example, Bruneau et al. (2022) aimed to promote peace and reintegration of FARC ex-combatants in Colombia. Therefore, the researchers measured a variety of outcome measures but focused on a few that included, most notably, participants’ support for the peace process in Colombia and their support for policies that aim to help ex-combatants reintegrate into Colombian society. On top of these outcomes, Bruneau et al. also measured relevant attitudinal and affective measures such as the perception that FARC ex-combatants are unwilling and unable to let go of violence, dehumanization, intergroup empathy, and prejudice.

Finally, in between those two ends of the spectrum, there are some instances in which a problem can be operationalized in several different concrete ways. For example, the Strengthening Democracy Challenge (2021) organizers are interested in strengthening U.S. democracy given rising levels of polarization and interpartisan prejudice (e.g., Iyengar et al., 2019). This rather abstract goal was then operationalized as three concrete outcomes (i.e., antidemocratic attitudes, support for partisan violence, and partisan animosity, which includes a behavioral measure). In other words, the tournament organizers were equally interested in promoting better interpartisan relations and combating the process of democratic norm erosion in the United States.

Next, in the vast majority of intervention tournaments, efficiency of interventions is determined by assessing whether the difference between each intervention and the control condition is statistically significant (below we elaborate on whether these analyses should be adjusted because of multiple comparisons) and by reporting the size of these effects. In the minority of cases, following the initial intervention tournament, and as we elaborate below, the efficiency of interventions is then assessed in a follow-up study that examines whether the effects persisted for a period of time and in replication studies that examine whether the effect is reliable (see e.g., Bruneau et al., 2022; Gallardo et al., 2022; Lai et al., 2014, 2016; Moore-Berg, Hameiri, Falk, & Bruneau, 2022).

It is common for researchers to compare the interventions to an empty control (i.e., no intervention) condition rather than to an active control (placebo) condition. This is because it is extremely difficult in the psychological sciences to come up with an active control condition that will not bias the results in any way (e.g., increase positive affect, cognitive flexibility), especially compared with several other, theoretically diverse, competing interventions. However, an empty control condition has two major limitations that an active control condition can address. Specifically, when an empty control is used, participants can realize that they are in the control condition, which can then affect how they respond to the outcome measures. Indeed, this is less of a concern when the outcomes are behavioral (e.g., in the case of getting flu vaccinations; Milkman et al., 2021) or implicit (e.g., in the case of the IAT; Lai et al., 2014, 2016). However, when the outcome measures are mostly self-reported, then the fact that participants are not blind to their condition can yield potential bias in the results because of demand characteristics (for a related discussion in medical research, see Freidlin et al., 2008). Second, using an active control can reduce potential selection bias (albeit, it cannot be completely eliminated; see Uschner et al., 2018). On the other hand, unlike an active control, an empty control prevents a potentially biased control condition and, as noted previously, requires no additional resources to deploy (i.e., does not require additional resources to develop the active control condition) and provides a potentially true baseline of attitudes and opinions.

Addressing Type I and II errors

Regardless of which type of control condition the interventions are compared with, multiple comparisons are being made, which requires the researchers to carefully consider how to address potential Type I (i.e., false positive) and Type II (i.e., false negative) errors. There are diverging views about whether researchers should include some (and what type of) statistical correction because multiple comparisons are being made. Some have argued that no correction is needed, especially in cases of intervention tournaments that examine distinct interventions and is exploratory, because the comparisons that are being made (between each intervention and control) are independent (Parker & Weir, 2020; Rubin, 2021). Others have argued that some type of correction is needed, especially in cases of intervention tournaments that examine iterations of the same intervention (which is mostly relevant to the in-house-developed intervention tournaments) and are confirmatory (Freidlin et al., 2008; Wason et al., 2014; Wason & Robertson, 2021). Note that intervention tournaments in the psychological sciences are more often exploratory than confirmatory.

Given these diverging views, we argue that to rule out potential Type I and Type II statistical errors, it is important for researchers to include multiple, preregistered, and statistically powered replication studies with new participants (e.g., Bruneau et al., 2022; Calanchini et al., 2021; Lai et al., 2014, 2016; Moore-Berg, Hameiri, Falk, & Bruneau, 2022). For instance, in Moore-Berg, Hameiri, Falk, and Bruneau (2022), the researchers initially conducted a 12-condition intervention tournament to examine the effectiveness of video interventions at reducing Islamophobia. From this initial intervention tournament, the authors identified three interventions that were most successful at decreasing support for punitive policies toward Muslims. They then conducted five follow-up preregistered and large-scale replication studies to rule out potential Type I and II errors. Likewise, Lai et al. (2014) conducted several replication studies following the initial intervention tournament and eventually tested each intervention 3.70 times on average (see also Axelrod, 1980a, 1980b). By replicating the results in preregistered and sufficiently powered studies, researchers can considerably reduce the concern that the effects they find in the intervention tournament are a mere fluke, which increases the results’ robustness. If replication studies are not feasible, researchers are advised to use various statistical techniques to account for multiple comparisons, such as reporting q values on top of the customary statistical indices (see Milkman et al., 2021; Wason & Robertson, 2021).

Underlying mechanisms of interventions

In addition to ruling out Type I and II errors, replication studies add further benefit of teasing out the underlying mechanism or mechanisms of the successful interventions identified in the tournament, especially when considering the curated and in-house-developed intervention tournaments. In most cases, researchers and practitioners will have some sense of the potential psychological mechanisms at play before the intervention tournament, as in Bruneau et al. (2018) and Moore-Berg, Hameiri, Falk, and Bruneau (2022; see also Calanchini et al., 2021). In other cases, researchers can tease apart a successful intervention after it is found to be successful to pinpoint the psychological mechanism from potential candidates. Both approaches have unique benefits associated with them and are not mutually exclusive. For instance, focusing on specific mechanisms before the intervention tournament can help the researchers during intervention curation. That is, the researchers might be interested in identifying interventions that increase empathy toward out-group members. Therefore, they might select only videos that appear to induce empathy on the basis of their intuitions. Then, the researchers might conduct a confirmatory study to ensure that empathy was the mechanism involved and/or tease apart additional psychological mechanisms that might be at play. However, there might be other cases in which researchers want to take a more exploratory approach to the intervention tournament and include, for example, interventions generally thought to improve implicit attitudes toward out-group members (see Lai et al., 2014). In this case, the researchers might include an assortment of promising interventions that appear to improve intergroup relations and focus only on identifying the mechanism after determining which intervention or interventions are most effective. For this more exploratory approach, the researchers might consider selecting several theoretical mechanisms that could drive the intervention’s effects a priori and then rule out which theoretical mechanism is most effective in subsequent follow-up studies.

Longitudinal intervention tournaments

As a final consideration for intervention-tournament design, some tournaments include a longitudinal component (e.g., Bruneau et al., 2022; Lai et al., 2016; Moore-Berg, Hameiri, Falk, & Bruneau, 2022). Including a longitudinal component as part of intervention-tournament design has several benefits. First, it can serve as another criterion for determining which intervention or interventions are most successful. For instance, Moore-Berg, Hameiri, Falk, and Bruneau (2022) administered the same questionnaire (without the interventions) to the same participants in the intervention tournament 1 month following the tournament. They considered an intervention to be successful only when it maintained its significant effects on the outcome during the 1-month follow-up study (see also Bruneau et al., 2022). Second, a longitudinal component can be another way to rule out Type I error and demand characteristics. By demonstrating that an effect lasts over time, researchers can have increased confidence that the effect was driven by the intervention itself rather than a statistical or methodological errors. Third, it can increase researchers’ confidence in the effectiveness of the intervention. Like all longitudinal studies, demonstrating that an effect lasts beyond the initial testing increases the robustness of the research. Unfortunately, only a small minority of all studies that assess the effectiveness of interventions include a longitudinal element (Paluck et al., 2021).

Conclusion

The goal of the current article was to increase the use of intervention tournaments in the psychological sciences by providing the pros and cons of this approach and practical considerations for psychological scientists who are interested in using it. We argue that intervention tournaments hold the potential to greatly improve scientific research—they allow for the efficiency of testing of multiple interventions at once, encourage the collaboration across research labs with diverse expertise and/or between academics and practitioners, and improve the external validity of research. Intervention tournaments also hold the potential to increase the rigor of applied research that assesses interventions that are designed and implemented by practitioners. However, we note that intervention tournaments are no panacea. Although it is an efficient approach, because it tests several interventions at once by means of a single control condition, a single study does require considerable resources to ensure sufficient statistical power, especially if researchers want to use a nationally representative sample or collaborate with organizations that have the capabilities and infrastructure to run such complicated studies in the field. Furthermore, and as mentioned above, the focus on identifying the more effective approach means that the intervention tournament does not inherently provide an answer to why the successful intervention or interventions was indeed the most effective.

As we have elaborated on in this article, there are several elements that we suggest researchers should incorporate in their research to use this approach effectively while also minimizing its limitations. These include, most notably, conducting replication studies that can address issues of statistical errors and shed light on potential psychological mechanisms and some limitations in the intervention-selection process. We argue that when this approach is used diligently, intervention tournaments complement existing approaches to intervention science by providing the opportunity for rigorous investigation of interventions against other interventions. This sort of rigorous testing is necessary to push the field of intervention science forward and establish theoretical bases for which interventions are most successful.