Research-of-1: My Esametric Vision

When you apply to academic faculty jobs, you have to write a research statement. In mine, I outlined my vision of a research career into a field of statistics for expanded n-of-1 studies.

But there was a problem: This field had no name.

I’d pitched my work on n-of-1 observational causal inference at various meetings, classes, workshops, and conferences. Throughout this formative time, I’d met a number of methodologists working in adjacent technical areas.

I began thinking, hrmm, maybe this field shouldn’t just cover statistical methods for n-of-1 trials and single-case experimental designs… Maybe it should also include a broader pool of techniques for analyzing newer single-subject observational studies—the kinds that had temporally dense data made available by today’s mobile or wearable devices and sensors. These modern data scenarios implicate additional areas like machine learning, functional data analysis, and actigraphy (among others).

I needed to be able to talk about statistics research fragmented across these broader disciplines under one catch-all term, one name to describe the set of approaches from each area that were being developed to study intra-individual cyclic patterns. But what?

Thankfully, 1.) I love puns and wordplay, and 2.) I’m proudly Filipino American. The field I’ve described is focused on studying patterns within one individual, often using biometrics. And the Tagalog Filipino word for “one” is isa (pronounced “ee-SA”). Furthermore, some might say I can be somewhat punbearable when it comes to portmanteaus…

I therefore dubbed this field esametry (pronounced “ee-sa-metry”). ✔️ I like the name because it also jives etymologically with the psychology term idiographic (which describes a study of just one person) without sounding too puntifical.

To help define the scope of esametry, in my research statement I identified the following broad areas of immediate need:

1. Continue adapting causal inference methods for n-of-1 observational study (N1OS) settings, following Daza (2018, 2019).
   1. Strengthen these novel methods by applying machine learning techniques to various modeling components.
2. Utilize population-based findings to reduce the number of viable causal structures over which to search for likely personalized causes, thereby easing the computational burden of causal-structure discovery.
   1. Adapt or create methods that “borrow strength” from similar individuals’ or population-level data (e.g., random-effects / mixed-effects models, meta-analysis).
   2. Work with health domain experts, as well as bioinformatics and computer science researchers.
   3. Use both designed datasets (e.g., from the U.S. Census Bureau) and convenience samples (i.e., “found data” like electronic health records and clinical -omics data).
3. Develop appropriate post-discovery (i.e., post-selection) inference approaches.
   1. In Daza (2018) and related work, a priori unknown treatment periods are first discovered (e.g., through changepoint detection or cluster analysis), and then subsequently assumed to be constant.
   2. This yields overly optimistic confidence limits.
4. Develop the scalability and modularity aspects of the average period treatment effect (APTE) framework of Daza (2018, 2019). (More on the APTE in a follow-up post. 🔍)
   1. Personalized analysis can involve causal structures at different scales.
   2. This work will further refine causal graphs and diagrams (e.g., directed acyclic graphs) used to probe multi-scale, multi-level hypotheses of causal mechanisms.
5. Investigate how to properly account for missing data in the n-of-1 setting.
   1. Single-subject self-tracked data are prone to haphazard reporting or recording (e.g., device non-wear time).
   2. Hence, they require careful assessment of the effect of missing data on analytical estimates or inferences.
6. Optimize retrieval and featurization of the wearable-device data used in N1OS studies.

My long-term goal is to build and grow an esametry lab or working group—and eventually, center—to coordinate these research areas. This lab will provide consultation to patients, health providers, statisticians, data scientists, and interested communities (e.g., ICN for N-of-1 Clinical Trials and SCEDs, Quantified Self) on how to use our personalized tools, as well as improve their adoption and utility through an iterative user-centered design approach.

But in the meantime, let’s start with this humble yet ambitious (humbitious?) blog. 😉

What do y’all think? Share your thoughts in the Comments below. ⬇️

If you’d like to get involved, or are already working in any of these areas (I’m lookin’ at you 👀) and would like to help define the field of esametry, email me at ericjdaza@statsof1.org, or message Stats-of-1 at twitter.com/statsof1.

And do let me know if you’d like to write a blog post sometime about your work—and how it advances esametry!