My April-June Recap
Last updated: Jul 20, 2021
I’m super stoked to announce that I’ll be presenting a new analytic method using wearable sensor data at the Joint Statistical Meetings (JSM) August 8-12 this year! JSM is “one of the largest statistical events in the world” with “more than 6,500 attendees from 52 countries. It is also one of the broadest, with topics ranging from statistical applications to methodology and theory to the expanding boundaries of statistics, such as analytics and data science.”
I’ve attended and presented at JSM numerous times on this ongoing n-of-1 methods development, and am happy to get the chance to keep the field updated. I originally presented this particular method at the Society of Behavioral Medicine (SBM) 42nd Annual Meeting this past April.
I call the method Model-Twin Randomization, or MoTR (“motor”). It’s a method for discovering or evaluating the recurring or chronic effects of a single person’s health behaviors. As an example application, my colleague Logan and I will start teasing apart how physical activity may have affected sleep duration (and vice versa) for each of us. (The example I presented at SBM addressed the converse question of how sleep duration may have affected physical activity.)
MoTR essentially applies statistical modeling (including fancy machine learning methods, if appropriate) to your wearable data to create your digital “model twin”, and then mimics an n-of-1 trial (i.e., a multi-period or multi-phase randomized controlled crossover trial on one person) to estimate possible effects by simulating randomization. Formally speaking, MoTR is an application of the g-formula (i.e., standardization, back-door adjustment) under serial interference. We’re working on a pre-print, which should be up in the next month or two. Stay tuned!
The idea is that you might miss something—might mistake correlation for causation—if you only examine how one variable relates to the other, without accounting for possible confounders. For example, if I notice that walking faster today correlates with longer sleep tonight, I might think walking faster on any given day causes me to sleep longer that night. But I’d be wrong if in fact walking faster throughout the week before yesterday causes me to both walk faster today AND sleep longer tonight—and that walking faster today doesn’t actually impact my sleep duration that much. So I could try to make myself walk faster on any given day, but I would fail to see any immediate impacts on my sleep that night.
I was also absolutely delighted to attend the KU Leuven Small is Beautiful (Again) Symposium! It was a very welcome crash-course for me on the psychology side of single-individual designs, or single-case experimental designs (SCEDs). I attended an excellent workshop on “Using the Mobile Application SCD-MVA for Masked Visual Analysis of Single-Case Studies” led by Professors John Ferron and Mariola Moeyaert.
It was super-exciting to finally get to virtually “meet” all these SCED thought leaders and pioneers! I was particularly happy to meet many of the quantitative methodologists present, including Professors Rumen Manolov and Patrick Onghena.
Recently, I was also featured on a data science podcast discussing n-of-1 causal inference. Check out Data & Science with Glen Wright Colopy for great discussions on modern health, data science, and statistics. Here's my contribution on my favorite topics: n-of-1, causal inference, and diversity: Eric Daza | N-of-1 Science & Causal Inference | Philosophy of Data Science
Attending conferences that span multiple disciplines is a great way to build interdisciplinary camaraderie and a shared knowledge base for single-individual methodology. Introducing folks from other subfields within each of these scientific domains also helps draw in collaborators with deep experience and expertise who otherwise wouldn’t have seen that their work might connect with ours. I hope my podcast appearance helped entice more statisticians, data scientists, and machine learners to join us at Stats-of-1!