I am currently recruiting graduate students at both the Masters and PhD level.

If you are interested in pursuing either a Masters or PhD in Statistics, and are interested in any of the research topics that I work on, please reach out. Even if you have a somewhat non-traditional background, but think that you would be a good fit, please get in touch with me.

These are funded opportunities with a lot of flexibility for you to grow and learn as a Statistician. Do not hesitate to reach out!

Back to 'My Thoughts'

My Research Philosophy Statement

job search
This document outlines my research experience, approach, and goals for the future.

Dylan Spicker


December 14, 2022

Questions of causality sit at the heart of scientific inquiry and progress. While historically causality has been inferred through the use of randomised trials and experiments, the rate of growth of modern observational data and the corresponding complexity necessitates the development of novel causal techniques in order to ensure continued scientific progress. I am a causal inference researcher with an interest in developing novel methods to accommodate the growing complexity of modern data. I focus on developing techniques which are rigorously justified while remaining widely accessible. I am strongly committed to interdisciplinary research, which helps to ensure that my independent research remains grounded and readily applicable. I have a record of productive, impactful research, capable of attracting competitive funding. My interests are in burgeoning areas, tackling modern statistical issues, with diverse applied utility, and strongly integrated interdisciplinary partnerships. This provides access to a wide variety of funding sources, and serves as a strong foundation for a publishable, fundable, independent research program.

Past Research

My past work focused on developing methods which account for the effects of measurement error and misclassification in the context of causal inference with dynamic treatment regimes (DTRs). DTRs provide a framework for capturing causal inference in a longitudinal setting, often applied for precision medicine, where we estimate the optimal sequence of treatments for individual patients. Measurement error is a pervasive issue, particularly in biostatistics, occurring whenever a variate of interest cannot be accurately observed. I provided the first account of the impacts of measurement error on optimal DTR estimation, and developed the first set of estimators to correct for these issues [1]. Additionally, I highlighted the shortcomings of ignoring patient nonadherence in optimal DTR estimation, and provided a locally efficient, computationally feasible estimation technique for the valid estimation of causal effects in these settings [2]. This work has resulted in a manuscript published in Statistics in Medicine, a top-tier Biostatistics journal, with a secondary article under consideration, has received competitive NSERC grant funding, and has since spawned additional contributions in the field [3, 4]. Beyond my work in modern causal inference, I have developed general measurement error corrections, leveraging both nonparametric [5] and robust [6] statistics, producing two additional manuscripts which are currently under review.

Current Research

I am continuing to expand methods for optimal DTR estimation with complex data, where I have begun work related to patient privacy in DTR estimation. Differential privacy is a standard which ensures that reported estimates do not reveal too much information regarding any individual in a study, and preserving this privacy is critically important for statistical analyses in precision medicine. I am also working on techniques for sample size and power calculations within the DTR framework, an important area missing from the current literature. Outside of work on DTRs, my postdoctoral research focuses on the analysis of data from respondent-driven sampling (RDS). RDS combines benefits from probability based sampling and snowball sampling, providing an effective technique for researchers studying hard-to-reach populations. It has correspondingly seen wide uptake by applied researchers. Despite its frequent application, existing techniques for RDS analysis focus primarily on mean and proportion estimation, in a cross-sectional setting. I am developing a set of causal estimators using data arising from RDS, techniques which are essential for answering scientific questions with RDS data. Additionally, I am extending both the causal estimators and the existing mean estimators to a longitudinal RDS setting. This work is currently funded with a competitive postdoctoral grant, issued by CANSSI, and is motivated through a partnership with the Engage study. Engage is an ongoing, multicenter RDS, operating out of Montreal, Toronto, and Vancouver, aiming to “inform HIV and STBBI prevention techniques in the population of gay, bi, and queer men, including trans men, and other men who have sex with men.”

Future Research

Going forward, I will continue to develop novel methods in causal inference for modern issues in real-world data. In the short-term I will study complex outcomes in the context of DTRs. In the DTR literature, it is common to assume that there is a single outcome to be optimised for all individuals. In practice there are often multiple, competing outcomes. A medical treatment may be thought to increase life expectancy, but this needs to be balanced against potential negative side effects. Existing work on competing outcomes in DTRs ignores the multivariate nature of the problem, opting instead for simplified combinations using utility functions. This assumes that utility is estimable or else can be constructed, and that future patients will balance their priorities similarly. My proposed strategy will provide more flexibility for combinations of competing outcomes, compared with univariate utility functions. DTRs are typically studied under the assumption that there is a known, finite number of decisions to make. There is some work considering infinite time horizon DTRs through the use of Markovian assumptions. While these assumptions are mathematically convenient, they can be oversimplified in real-world scenarios, particularly in medical settings where delayed treatment effects are common. I will consider a joint modelling approach, treating the number of decisions as a random quantity in addition to the outcomes of interest. This joint modelling approach will borrow from standard longitudinal and survival analysis, and will provide a more realistic set of assumptions for real-world decision making. In the longer term, I will develop tools for causal inference on complex, modern data types. For instance, I will develop causal methodologies for both network and functional data. Both network and functional data are becoming more prominent, and each represents an important form of “big data”. Alongside the associational techniques used to study these large data sets, causal methods will be required to answer scientific questions which arise. This is particularly important as observational, rather than experimental, data are driving the increase in the availability of network and functional data. Causal network analysis has received some attention in the literature, but has remained relatively simple both in terms of associational structures and causal estimands (methods tend to assume only local interference patterns on networks in a way which often is not realistic). Functional data, on the other hand, have received comparatively little causal attention, despite the potential applied utility (for instance, with fMRI data).

Existing and Planned Collaborative Research

Beyond my impactful methodological developments, I have a strong focus on conducting interdisciplinary research. The purpose of this is two-fold: first, working with applied researchers helps to discern important methodological needs, ensuring that my independent research remains relevant. Second, the primary purpose of statistical methodology is for the advancement of scientific knowledge. Working on interdisciplinary research facilitates the theoretically valid application of cutting-edge statistical techniques, enhancing the pursuit of scientific knowledge. An example of this can be seen during my doctoral studies, where I established an ongoing research project with nutritional scientists, computer scientists, and statisticians. Our research aims to study the impacts of diet on health outcomes, while taking into account the complexities of dietary patterns (including combination effects which are historically ignored), using deep learning techniques. This ongoing project involves statistical methodology, algorithmic development, and nutritional science. In addition, during my postdoctoral studies, my work on RDS is directly informed by a close working relationship with medical and public health researchers from the aforementioned Engage study. This partnership ensures that the methodological developments are necessary, and will see applied use. Moving forward, I intend to continue initiating and participating in relevant interdisciplinary work, particularly in the health sciences. This provides exciting avenues for collaboration, a wider array of funding sources, and helps to ensure uptake of methodological developments. Several of my short-term projects provide opportunities for interdisciplinary research in personalised medicine, public health, and medical imaging. In addition to health applications, the possibility of industry collaboration exists, particularly related to functional and network data. Functional data is commonly observed in finance and speech recognition settings, while network data naturally arises in social networks and logistics tasks.


[1] Spicker, Dylan, and Wallace, Michael P. Measurement error and precision medicine: Error-prone tailoring covariates in dynamic treatment regimes. Statistics in Medicine. 2020; 39: 3732– 3755. https://doi.org/10.1002/sim.8690

[2] Spicker , Dylan. Generalizations to Corrections of Measurement Error Effects for Dynamic Treatment Regimes. UWSpace. 2022. http://hdl.handle.net/10012/18581.

[3] Wallace, Michael. P. Measurement error and precision medicine. In T. Cai, B. Chakraborty, E. Laber, E. Moodie, and M. van der Laan, editors, Handbook of Statistical Methods for Precision Medicine [Accepted], Handbooks of Modern Statistical Methods. Chapman and Hall/CRC, Boca Raton Florida. 2022

[4] Liu, Dan. Regression-based Methods for Dynamic Treatment Regimes with Mismeasured Covariates or Misclassified Response. 2022.

[5] Spicker, Dylan, Wallace, Michael P., and Yi, Grace Y. Nonparametric Simulation Extrapolation for Measurement Error Models. arXiv preprint arXiv:2111.02863. 2021.

[6] Spicker, Dylan, Wallace, Michael P., and Yi, Grace Y. Generalizations to Corrections for the Effects of Measurement Error in Approximately Consistent Methodologies. arXiv preprint arXiv:2106.07401. 2021.