A personal take on science and society

World view
By Balazs Aczel

Let the data talk: embrace

exploratory research
Open-ended research is essential to my hypothesis-testing analyses properly.
building solid hypotheses — without it, For example, how could a researcher tackling the aggres-
sion question know, without exploration, which of many
even the best-planned analyses can fail. Institutions variables — age, gender, socio-economic background and
must not

A
more — might be linked to behavioural change, and how?
ttending a psychology conference this regard Similarly, in my own work, I’ve found myself asking: should
month, I was struck by an unsettling trend in
exploratory I have devised my statistical model differently? Would a dif-
social-science research. The abstract book ferent analysis bring different conclusions? Did I miss some-
revealed that nearly all of the 1,500 posters studies as thing more interesting in my data? I felt unable to let my data
and talks related to some kind of confirmatory lower-league speak for themselves. I should have done more exploration
work — using statistical analyses to test an existing hypothe- research.” before committing to my confirmatory analysis.
sis. Hardly any reported the sort of open-ended exploratory Yet, despite the importance of exploratory studies, they
research needed to come up with those hypotheses. are not valued highly enough. Hundreds of journals pro-
In my view, it’s crucial for the social sciences to embrace mote registered reports — a way to preregister confirma-
a research culture that balances hypothesis testing with tory research with the journal — yet exploratory science is
hypothesis generation, embracing exploratory studies rarely endorsed. And in the current reproducibility drive,
alongside confirmatory ones. with questions raised about how to make the social sciences
To understand the difference between exploratory and more rigorous, robust and reproducible, most solutions
confirmatory studies, consider a research group that is have involved improving how we test our hypotheses, with
interested in links between children who watch violent little thought given to exploratory work.
films and aggressive behaviour. In an exploratory study, Many researchers are not trained to conduct exploratory
the group might ask parents about the children’s aggres- studies. What’s more, many feel uneasy about doing such
sive behaviour and collect information about household work. One study found that exploratory research is viewed
habits, individual characteristics and media content that as ‘less scientific’ than confirmatory research (H. K. Collins
could influence the children. The authors would inspect the et al. Organ. Behav. Hum. Decis. Process. 164, 179–191; 2021),
data in many ways, looking for any patterns. The patterns probably owing to the emphasis the latter has received.
that emerge, alongside theoretical background knowl- To help turn the tide, journals should make it clear that
edge, could then be used to build specific predictions — they welcome exploratory studies. I know of a handful that
for instance, that parents’ explanation of violent content already advertise exploratory reports — articles that can
might decrease media-induced violence in adolescents, include not just exploratory data analyses, but also descrip-
but not younger children. tive studies, or those that add constraints to existing
Confirmatory studies can seek to support or question theories. More journals should embrace this idea.
these hypotheses using a fresh data set. To conduct a Researchers undertaking exploratory work should know
confirmatory study properly, best practice dictates that that this doesn’t mean they can be lax about scientific rig-
researchers outline their analysis plan before beginning our. The criteria that peer reviewers should consider when
their work, often through preregistration — recording it evaluating exploratory studies are similar to those for con-
in a public repository. Preregistration helps to prevent firmatory studies: the rationale for a particular study; its
researchers from constructing their hypotheses and potential value to the field; whether its data are presented
analyses after peeking at their data. In the example above, clearly; its theoretical relevance; and the transparency of
for instance, if a group’s analyses do not confirm the gen- the report. The aim of peer review should be to ascertain
eral hypothesis, the researchers might be tempted to tweak that an exploratory study provides well-founded and spec-
the type of analysis they perform until it shows a statisti- ified predictions for future confirmatory studies to test.
cally significant effect for, say, early adolescent boys, after Institutions must not regard exploratory studies as
controlling for family income. This broadly condemned lower-league research, either. Hypothesis testing, as the
practice opens the door to false-positive findings. core of social-sciences research, is often front and centre in
Like many in the social sciences, I was trained to think Balazs Aczel is undergraduate courses. But students of the social sciences
of hypothesis testing as the main job of a researcher, with a professor of should also be taught exploratory techniques and theory
exploration hardly worthy of publication. But I changed psychology at Eötvös building throughout their degrees.
my mind as I began to lose faith in my own ability to pre- Loránd University, When we are unsure about what we are after, we should
G. Z. HORVATH

register my studies properly. I wholeheartedly believe that Budapest, Hungary. stop and explore. Instead of committing ourselves to arbi-
preregistration is crucial to transparent science. But often, e-mail: aczel.balazs@ trary specifications, we could let the data talk first. It could
I would find that I hadn’t done enough exploration to plan ppk.elte.hu be the start of a healthier relationship.

788 | Nature | Vol 635 | 28 November 2024