Janneke de Boer is a Marie Curie doctoral fellow within the Health CASCADE project. Her research focuses on which co-creation methods are most effective when working with teachers, school staff and parents.
Lea Rahel Delfmann is a Marie Curie doctoral fellow within the Health CASCADE project. Her research focuses on scaling an existing sleep intervention with the aim to preserve ownership among adolescents.
Janneke and Lea are looking at how to develop, implement, and evaluate co-created interventions in schools to promote healthy sleeping behaviour. As part of this process, they have been reflecting on how to iterate and scale these interventions. In this blog, they argue that co-creation moves in shades of grey – just like Bayesian statistics.
Part I: Frequentists’ thinking versus Bayesians’ thinking
“Today’s posterior is tomorrow’s prior”
In other words, the answer to our question today might be a question itself tomorrow. This idea of sequential updating of knowledge during research is linked to Bayesian statistics, an approach that is slightly different from the more conventional Frequentist approach.
So, what are the differences between Frequentists’ and Bayesians’ thinking? According to Dienes:
- Frequentists would state that the probability that a person is dead (the data) given that a shark has bitten their head off (the theory) is 1.
- Bayesians would state that the probability that a shark has bitten a person’s head off (the theory) given that the person is dead (the data) is close to 0.
The Frequentists’ statement does not say anything about the probability that the theory is true, but about the probability of observing the data, given that the theory is true. Using a Bayesian approach to the shark attack allows to say something about the probability that the theory is true given the data. With this, we can update the plausibility of our theory with new data.
Part II: A Bayesian interpretation of co-creation
We think that the Bayesian approach has something in common with the co-creation of interventions. Within Health CASCADE, we are interested in applying a robust methodology, and building blocks through appropriate methods that make our interventions successful, not only in whether our interventions are successful. We could say that we are interested in updating our theory, instead of just finding a significant intervention effect. Eventually, it is this flexible updating of our “theory” which makes the generalization of interventions possible.
Part III: Generalization of co-created interventions
Let’s talk more about generalization. By nature, co-created interventions only apply to a very specific group of people, namely the group of people who created them. Take for example a healthy nap intervention for toddlers in the ‘ladybug’ group of your niece’s kindergarten in Ghent. If we’d like to make this healthy nap intervention useful for all toddlers in all kindergartens in Belgium, Europe, or the world (if we want to set the bar very high), we need to scale the intervention. To do that, we might take the intervention to another setting and adapt it in collaboration with another group of toddlers, like the ‘cockroaches’ group in your best friend’s kindergarten in Brussels. In this process, our main interest is not to develop the perfect one-fits-all intervention, but to find the building blocks in the process that made the intervention work. In Bayesian terms, the intervention developed in the ‘ladybug’ group might be the prior, updated with data from the ‘cockroaches’ group. The updated intervention would then be the posterior until we go on looking for another group of toddlers in another kindergarten. During this iterative process, the building blocks that work for all toddlers will receive a boost in plausibility. The ones that don’t will suffer a decline. By focusing on the building blocks in the co-creation process rather than on the effect, co-created interventions become sustainable and applicable in various settings and contexts. This relates back to our starting point – co-creation is very similar to Bayesian statistics when it comes to the question of what works and what does not, or what is right and what is wrong. The answer is for now, that this has to be updated.
