\(\DeclareMathOperator{\expit}{expit}\) \(\DeclareMathOperator{\logit}{logit}\) \(\DeclareMathOperator*{\argmin}{\arg\!\min}\) \(\newcommand{\indep}{\perp\!\!\!\perp}\) \(\newcommand{\coloneqq}{\mathrel{=}}\) \(\newcommand{\R}{\mathbb{R}}\) \(\newcommand{\E}{\mathbb{E}}\) \(\newcommand{\M}{\mathcal{M}}\) \(\renewcommand{\P}{\mathbb{P}}\) \(\newcommand{\I}{\mathbb{I}}\) \(\newcommand{\1}{\mathbbm{1}}\)

Welcome

Targeted Learning in R: Causal Data Science with the tlverse Software Ecosystem is an fully reproducible, open source, electronic handbook for applying Targeted Learning methodology in practice using the software stack provided by the tlverse ecosystem. This work is a draft phase and is publicly available to solicit input from the community. To view or contribute, visit the GitHub repository.


Outline

The contents of this handbook are meant to serve as a reference guide for both applied research and for the teaching of short courses illustrating successful applications of the Targeted Learning statistical paradigm. Each section introduces a set of distinct causal inference questions, often motivated by a case study, alongside statistical methodology and open source software for assessing the scientific (causal) claim of interest. The set of materials currently includes

What this book is not

This book does not focus on providing in-depth technically sophisticated descriptions of modern statistical methodology or recent advancements in Targeted Learning. Instead, the goal is to convey key details of these state-of-the-art statistical techniques in a manner that is clear, complete, and intuitive, while simultaneously avoiding the cognitive burden carried by extraneous details (e.g., mathematically niche theoretical arguments). Our aim is for the presentations herein to serve as a coherent reference for researchers – applied methodologists and domain specialists alike – that empower them to deploy the central statistical tools of Targeted Learning in a manner efficient for their scientific pursuits. For a mathematically sophisticated treatment of some of these topics, inclusive of in-depth technical details, in the field of Targeted Learning, the interested reader is invited to consult van der Laan and Rose (2011) and van der Laan and Rose (2018), among numerous other works, as appropriate. The primary literature in causal inference, machine learning, and non/semi-parametric statistical theory include many of the most recent advances in Targeted Learning and related areas. For background in causal inference, Hernán and Robins (2022) serves as an introductory modern reference.

About the authors

Mark van der Laan

Mark van der Laan is Professor of Biostatistics and of Statistics at UC Berkeley and co-director of UC Berkeley’s Center for Targeted Machine Learning and Causal Inference. His research interests include statistical methods in computational biology, survival analysis, censored data, adaptive designs, targeted maximum likelihood estimation, causal inference, data-adaptive loss-based learning, and multiple testing. His research group developed loss-based super learning in semiparametric models, based on cross-validation, as a generic optimal tool for the estimation of infinite-dimensional parameters, such as nonparametric density estimation and prediction with both censored and uncensored data. Building on this work, his research group developed targeted maximum likelihood estimation for a target parameter of the data-generating distribution in arbitrary semiparametric and nonparametric models, as a generic optimal methodology for statistical and causal inference. Since mid-2017, Mark’s group has focused in part on the development of a centralized, principled set of software tools for targeted learning, the tlverse.

Jeremy Coyle

Jeremy Coyle, PhD, is a consulting data scientist and statistical programmer, currently leading the software development effort that has produced the tlverse ecosystem of R packages and related software tools. Jeremy earned his PhD in Biostatistics from UC Berkeley in 2017, primarily under the supervision of Alan Hubbard.

Nima Hejazi

Nima Hejazi is Assistant Professor of Biostatistics at the Harvard T.H. Chan School of Public Health. He obtained his PhD in Biostatistics at UC Berkeley working with Mark van der Laan and Alan Hubbard, and afterwards held an NSF Mathematical Sciences Postdoctoral Research Fellowship. Nima’s major research interests blend causal inference, machine learning, non/semi-parametric inference, and computational statistics; areas of recent emphasis have included nonparametric causal mediation analysis, efficient estimation under biased sampling designs, and sieve estimation with machine learning. His methodological work is motivated principally by scientific collaborations in clinical trials, infectious disease epidemiology, and computational biology. Nima is also passionate about statistical computing and open source software design for statistical data science.

Ivana Malenica

Ivana Malenica is a Postdoctoral Researcher in the Department of Statistics at Harvard and a Wojcicki and Troper Data Science Fellow at the Harvard Data Science Initiative. She obtained her PhD in Biostatistics at UC Berkeley working with Mark van der Laan, where she was a Berkeley Institute for Data Science and a NIH Biomedical Big Data Fellow. Her research interests span non/semi-parametric theory, causal inference and machine learning, with emphasis on personalized health and dependent settings. Most of her current work involves causal inference with time and network dependence, online learning, optimal individualized treatment, reinforcement learning, and adaptive sequential designs.

Rachael Phillips

Rachael Phillips is a PhD student in biostatistics, advised by Alan Hubbard and Mark van der Laan. She has an MA in Biostatistics, BS in Biology, and BA in Mathematics. As a student of targeted learning, Rachael integrates causal inference, machine learning, and statistical theory to answer causal questions with statistical confidence. She is motivated by issues arising in healthcare, and is especially interested in clinical algorithm frameworks and guidelines. Related to to this, she is also interested in experimental design; human-computer interaction; statistical analysis pre-specification, automation, and reproducibility; and open-source software.

Alan Hubbard

Alan Hubbard is Professor of Biostatistics at UC Berkeley, current chair of the Division of Biostatistics of the UC Berkeley School of Public Health, head of the data analytics core of UC Berkeley’s SuperFund research program, and co-director of UC Berkeley’s Center for Targeted Machine Learning and Causal Inference. His current research interests include causal inference, variable importance analysis, statistical machine learning, estimation of and inference for data-adaptive statistical target parameters, and targeted minimum loss-based estimation. Research in his group is generally motivated by applications to problems in computational biology, epidemiology, and precision medicine.

Reproduciblity

The tlverse software ecosystem is a growing collection of packages, several of which are quite early on in the software lifecycle. The team does its best to maintain backwards compatibility. Once this work reaches completion, the specific versions of the tlverse packages used will be archived and tagged to produce it.

This book was written using bookdown, and the complete source is available on GitHub. This version of the book was built with R version 4.2.0 (2022-04-22), pandoc version 2.7.3, and the following packages:

package version source
bookdown 0.26.3 Github (rstudio/bookdown@169c43b6bb95213f2af63a95acd4e977a58a3e1f)
bslib 0.3.1 CRAN (R 4.2.0)
dagitty 0.3-1 CRAN (R 4.2.0)
data.table 1.14.2 CRAN (R 4.2.0)
delayed 0.3.0 CRAN (R 4.2.0)
downlit 0.4.0 CRAN (R 4.2.0)
dplyr 1.0.9 CRAN (R 4.2.0)
forecast 8.16 CRAN (R 4.2.0)
future 1.26.1 CRAN (R 4.2.0)
ggdag 0.2.4 CRAN (R 4.2.0)
ggfortify 0.4.14 CRAN (R 4.2.0)
ggplot2 3.3.6 CRAN (R 4.2.0)
kableExtra 1.3.4 CRAN (R 4.2.0)
knitr 1.39 CRAN (R 4.2.0)
mvtnorm 1.1-3 CRAN (R 4.2.0)
origami 1.0.5 Github (tlverse/origami@e1b8fe6f5e75fff1d48eed115bb81475c9bd506e)
randomForest 4.7-1.1 CRAN (R 4.2.0)
readr 2.1.2 CRAN (R 4.2.0)
rmarkdown 2.14 CRAN (R 4.2.0)
skimr 2.1.4 CRAN (R 4.2.0)
sl3 1.4.5 Github (tlverse/sl3@de445c210eefa5aa9dd4c0d1fab8126f0d7c5eeb)
stringr 1.4.0 CRAN (R 4.2.0)
tibble 3.1.7 CRAN (R 4.2.0)
tidyr 1.2.0 CRAN (R 4.2.0)
tmle3 0.2.0 Github (tlverse/tmle3@ed72f8a20e64c914ab25ffe015d865f7a9963d27)
tmle3mediate 0.0.3 Github (tlverse/tmle3mediate@70d1151c4adb54d044f355d06d07bcaeb7f8ae07)
tmle3mopttx 1.0.0 Github (tlverse/tmle3mopttx@c8c675f051bc5ee6d51fa535fe6dc80791d4d1b7)
tmle3shift 0.2.0 Github (tlverse/tmle3shift@4ed52b50af501a5fa2e6257b568d17fd485d3f42)

Learning resources

To effectively utilize this handbook, the reader need not be a fully trained statistician to begin understanding and applying these methods. However, it is highly recommended for the reader to have an understanding of basic statistical concepts such as confounding, probability distributions, confidence intervals, hypothesis tests, and regression. Advanced knowledge of mathematical statistics may be useful but is not necessary. Familiarity with the R programming language will be essential. We also recommend an understanding of introductory causal inference.

For learning the R programming language we recommend the following (free) introductory resources:

For a general, modern introduction to causal inference, we recommend

Feel free to suggest a resource!

Want to help?

Any feedback on the book is very welcome. Feel free to open an issue, or to make a Pull Request if you spot a typo.