Table 8.

Recommendations for researchers and policymakers

Recommended future applications for science and policy		Summary	Example/application
	Formulate testable claims	Many prospective articles did not specify the directions of anticipated effects or identify causal factors. Others proposed complex components that were too nuanced to test. Meaningful precision in specifying claims and hypotheses would make it easier for practitioners to trial and incorporate evidence into policies	Prospective recommendations often indicated behaviours that might be ‘important’, but this creates an arbitrary standard as anything could conceivably be determined important or unimportant depending on the desired interpretation
	Study non-WEIRD populations	Recommendations made on the basis of evidence from WEIRD samples should not be uncritically generalized beyond these populations	Countries may need specific, local strategies that differ from those implemented in countries where much of the research has been done, because of differences in socioeconomic, political and macroeconomic conditions
	Think inside-out and outside-in	Claims from Van Bavel et al. primarily focused on the likely relevance of beliefs, perceptions, identity and other latent constructs. Research during pandemics should also focus on knowing what behaviours are most critical and the best ways to promote them, as well as identify interventions that consider structural contexts, not only psychological constructs	Early-stage recommendations from behavioural science in future pandemics should cover latent constructs (for example, identity, perceptions and norms), objective behaviours (for example, getting vaccinated and wearing a mask) and systemic factors (for example, access to the Internet, availability of healthcare and local legislation)
	Avoid the ‘streetlight effect’ by researching what matters, not just what is easy	Research during the pandemic often focused on what or who was easy to study rather than on what was most pressing for public health or who was most affected by the pandemic. Behavioural scientists should collaborate with practitioners to develop ways to make sure research resources are deployed where they will have the greatest impact. The most-studied populations in the pandemic were those with easy, stable access to the Internet. The most widely studied topics appeared to be those that could be tested through online surveys	Hundreds of studies on messaging were carried out online in WEIRD populations, but these did not demonstrate overall higher impact findings (in this review, at least)
	Test your assertions or programme evaluation	Many articles from 2020 made strong predictions that were not tested. The lack of clear validation or rejection of potentially influential policy interventions may allow some ineffective policy interventions to crowd out or divert resources from potentially more effective policy interventions	There was very little research studying whether the term ‘physical distancing’ would have more positive effects than ‘social distancing’
	Amplify according to evidence	The interventions getting the most attention were not necessarily those best supported by the most evidence. For example, correlations based on observational data were interpreted as if they were causal	Handwashing was widely promoted as a strategy for stopping the spread of COVID-19, yet study effects were small to null, particularly compared with masking, isolation, distancing and vaccines
	Precision, error, uncertainty and reality checks are always important	During the pandemic, public health agencies relied heavily on mathematical models with implausible assumptions about human behaviour. Behavioural scientists can improve these models by focusing on risk perception, innumeracy, noise, uncertainty and barriers to health behaviours (for example, access and costs)	When building models predicting behaviours such as vaccine uptake and isolation, factor in deviations from expectations based on practical, psychological factors
	Highlight null results	There is as much to learn from effective as ineffective interventions. Failures and backfire effects warrant more visibility to reduce attention to and spending on harmful or wasteful policies	A field experiment showing no significant effect of geo-targeted vaccine lotteries received very little coverage and influence on public policy
	Consider larger context	Research findings may vary depending on national, subnational and other local settings. Translating to policy interventions may require substantial adaptations to replicate effects	Appeal to national identity in liberal versus authoritarian regimes will probably differ in its behavioural consequences and ethical implications
	Do not overcommit too early	Although there are understandable pressures to issue guidelines quickly, establishing a policy position on poor or little evidence can lead to greater costs in the long term	Early guidelines in some countries suggested that wearing masks would not minimize COVID-19, but subsequent evidence has pointed to their effectiveness

Recommendations are ordered from primarily scientific to primarily policy, although to some extent, each recommendation applies to both.