Cooking Up Better Habits: What the Research Says About Designing Your Own Experiments
A synthesis of 3 studies on designing effective personal experiments — what actually works, what doesn't, and how to test it yourself.
Don't Expect Miracles: Lifestyle Changes Lead to Modest, But Real, Results
If you're "cooking up" a personal experiment to improve your health or habits, it's easy to get swept up in the promise of dramatic transformations. But research suggests a more grounded reality: lifestyle interventions, even when structured, often yield modest average effects. For instance, a Cochrane meta-analysis of 15 randomized controlled trials (RCTs) found that lifestyle changes in women with Polycystic Ovary Syndrome (PCOS) led to an average body weight reduction of just -1.68 kg. This isn't a headline-grabbing number, but it’s a concrete, measurable shift that, when understood in context, can guide your expectations and design choices for your own n=1 journey.
What the research actually shows
When we talk about "lifestyle changes," what are we actually doing? A crucial step in understanding what works comes from the Behaviour Change Technique Taxonomy version 1 (BCTTv1). This extensive 3-year project developed a standardized "menu" of 93 distinct behaviour change techniques (BCTs), such as goal-setting, self-monitoring, and social support. The research, involving approximately 400 experts and practitioners, demonstrated that trained coders could reliably identify these BCTs in intervention descriptions, with intercoder reliability (PABAK) often exceeding 0.70 (considered strong agreement). This taxonomy doesn't tell us which BCTs are most effective, but it provides a shared, precise language for describing the "active ingredients" of any intervention, allowing for clearer communication and more replicable experiments. For anyone designing a personal experiment, this means you can articulate exactly what you're trying to change and how.
Applying these "ingredients" to specific health conditions, a Cochrane meta-analysis investigated lifestyle interventions in 498 women with PCOS across 15 RCTs. These interventions included exercise alone, combined diet + exercise + behavioural coaching, or behavioural interventions. The findings, while based on low-quality evidence, revealed consistent if modest effects: participants experienced an average reduction in free androgen index (FAI) of -1.11 units, a decrease in body weight of -1.68 kg, and a drop in BMI of -0.34 kg/m². While these numbers might seem small, they represent real biological and anthropometric changes observed across diverse settings (USA, Australia, Europe, Iran, Turkey). This synthesis underscores that even broad lifestyle changes can move the needle, but the magnitude of change might be less dramatic than often portrayed.
Beyond specific outcomes, the feasibility of implementing interventions is paramount, especially in challenging environments. A study in Turbo, Colombia, explored a multi-component psychosocial intervention for older adults with unmet mental health needs. While a single-arm feasibility study, meaning it didn't compare the intervention to a control group, it focused on whether the program was deliverable and acceptable. Such studies, typically involving 30-80 participants, are critical for understanding the practicalities of an intervention in a specific context – including recruitment rates, retention, and fidelity of delivery. For your own experiment, this highlights the importance of not just tracking outcomes, but also the process: how consistently did you apply your intervention? Was it sustainable? Was it acceptable to you?
The nuance most people miss
One common oversight is equating "lifestyle change" with a magic bullet. The PCOS meta-analysis, for example, grouped various interventions – from exercise alone to combined diet, exercise, and behavioural coaching. This broad categorization means it's difficult to pinpoint which specific "recipe" of BCTs yielded the observed modest effects. Furthermore, the evidence quality was low, suggesting that individual responses could vary widely. What works for one person in a study