Big term science

Can some of science’s biggest questions simply not be answered without redefining how research is done? This is what motivated the founders of the ManyBabies Consortium, a network of 450 scientists from more than 200 institutions that pools resources to conduct large-scale research in infant development.

Our position is: there are some big scientific questions that simply can’t be answered without changing the way research is currently done. Large-scale team science is the answer. And it was this understanding that motivated the researchers who later established ManyBabies: a grass-roots network of researchers from across 200 institutions in 25 different countries, who collect and pool resources to conduct cross-cultural studies on infant development (see ref 1).

The ManyBabies Collaboration launched in March 2017. Since then, it has grown to include some 450 researchers from more than 200 institutions, who have pooled resources to complete massive studies on infant development. The consortium is delivering results about how babies learn across cultures and languages, and at different ages. But bigger questions remain.

How might science benefit from a rethink of how research is done? It’s a question central to the work of the ManyBabies Consortium, the umbrella organization for large-scale collaborations on infant and child development. Inquiries into fields like artificial intelligence, public policy and education can be informed by understanding the factors that affect human learning, an area in which infants are uniquely robust. Yet these same studies may require too many resources for a single lab to complete alone. The consortium provides a model for meeting this need.

It’s likely that none of these factors is individually responsible for infants’ attention — they are simply too complex. Instead, researchers thought that the best way to understand such learning was to carry out many experiments on many babies, each focused on a different condition, with the realized effect calculated across all of the studies combined. The ManyBabies Consortium was born in 2016, as a platform for doing just that.

Although cognitive scientists are making progress, many question whether we can find the ‘true’ answers to these questions under the current research model. In this paper, team members of the ManyBabies Consortium describe their network as an alternative model to answer questions that are too big for a single researcher or lab to tackle alone.

The ManyBabies Consortium aims to answer these questions by pooling data about infants from labs all over the world.

Building from a pilot meeting in 2015, the ManyBabies consortium has since conducted “precompetitive studies” across 200+ institutions on a number of topics: behaviour (including perceptions of speech, categorization and learning) and genetics.

Understanding how infants learn requires the rigorous examination of these complexities, but many researchers worry that the field’s reliance on small, non-randomized studies means that conclusions will be flawed by publication bias and inconsistent definitions.

Fields: Early Childhood development, Biology, Psychology and Comparative Cognition

What captures an infant’s attention? The probability depends on presentation (by a mother or stranger), the child’s previous experiences with mammals, what else is present alongside the rabbit, and more. To control for each of these factors, it would take thousands of hours of an infant’s time — a prohibitively expensive proposition. But researchers in the ManyBabies Consortium are pooling resources to complete massive studies on infant development under standard conditions.

The goal of ManyBabies Consortium — manybabies.org — is to build a research community that enables scientists, who would otherwise be competitors, to share resources and test theories on infant learning together.

But the hundreds of variables that could play a role, combined with the difficult nature of eliciting appropriate behaviours from infants, means that each hypothesis-testing trial takes weeks or months to complete and that very little is known about attention. To date, the field has conducted only 20 of these studies. Yet when researchers in the consortium completed 39 such trials in two years at an average cost of US$20,000 per study, they found that the existing literature on infant attention was unreliable: the novel results differed from almost all previous findings2. They concluded that important questions about human learning can be answered only through large collaborative networks.

Infants listen and learn from the very beginning of their lives. Yet research on infant learning is staggeringly difficult and expensive to conduct, as it typically involves gathering data from only a handful of infants each year. What’s more, many of the methods used in developmental psychology simply aren’t well suited to reliable findings in small samples. The Big-Team Science manifesto begs researchers (especially grant funders) to join existing large-scale teams, or create new ones, in order to answer some of our biggest questions.

As researchers interested in infant learning, we were divided between approaches. Should we concentrate on collecting data from a very large number of babies at a single institution, or should we collect data from a single baby tested at different times? How big was big enough?

Many researchers would like to answer this question, but it’s a huge study — one that takes months and costs thousands of dollars per child. Even by joining forces on a single task, the number of children studied is limited, the results inconsistent, and the quality of interpretations constrained. The Consortium is building up big science in psychology by collecting high-quality data from many labs around the world, creating new methods for working together, and attracting fresh talent. Any time you’re ready, we welcome you to join us.

This raises the question of how researchers can possibly capture the full spectrum of influences on infants’ attention. Typically, many such factors are not considered at all — for example, because researchers lack the resources needed to control and manipulate them (see ‘Big science for little people’). Even when factors are considered, their combined effects tend to be under-researched. As a result, most studies end up overlooking whole classes of important influences — so that much of what we think we know about infants does not correctly generalize across environments and situations.

In the past, infants were studied one at a time (the traditional US method), or in groups of up to 40 (the traditional European approach). This made it infeasible to hold things constant, and therefore impossible to get at the answer to some of the questions that psychologists like Woodward wanted to ask. So they decided do something radical — to bring together a large group of labs from around the world and run experiments over many years. In doing so, they developed methods for creating high-quality studies on a massive scale and for distilling those findings into coherent scientific advice.

No single lab will have the resources necessary to explore the many possible combinations of these factors. Yet, it is this diversity of scenario that stimulates infants’ curiosity and curiosity-driven attention — a key ingredient to successful learning — which has been supported by findings from psychology3 and neuroscience4,5. Even in cases where you could find enough participants, an underpowered study could prevent you from identifying the factors that really drive infant attention.

It is so difficult to isolate the effects of any one factor on infants’ attention, or any other cognitive skill, that a single researcher could not hope to characterize the complex whole.

In a new paper in the journal