A Conversation on Causality at Our Table (EN)
Brewing coffee, exchanging words. As steam rises, time drifts, and the world of statistics opens little by little.
This site grew from small coffee-chat conversations between a clinician daughter and her statistician father. Questions that surface in research, numbers that catch while reading a paper, and occasional glimpses of a researcher’s stance—the things that truly matter tend to lie scattered quietly across everyday life, clinical practice, causality, data, cognition, and language.
Over coffee, their light exchanges pick up some of these fragments, one by one—woven into a single story, as if careful not to disturb their totality.
Told in words warm enough to hold on a winter morning, the narrative moves with a spiral of quiet foreshadowing. Mathematics and medicine appear along the way, but take only what you need. Some episodes include short R scripts—small experiments you can rerun, so the ideas remain reproducible and tangible. If you follow the road all the way to the morning a paper is finally ready to submit, scattered meanings begin to connect—somewhere, into something whole.
How to begin
This story follows a single clinical study, moving in the order of research hypotheses → statistics → causality → language. If you are not sure where to start, these three episodes are good entry points:
- The first conversation and a research hypothesis (Study Design I)
- Clinical trials and p-values (Frequentist Thinking I),
- causal inference in research (Adjusting for Bias I)
Each episode can be read on its own—feel free to choose whichever layer interests you.
If you came here from an R post, you may want the code first. These scripts are meant to reproduce the figures and ideas, not to teach programming.
- Study Design: study-design.R
- Frequentist Thinking / Experiments: frequentist.R
- Effects and Time: effects.R
- Adjusting for Bias: logistic-regression.R
Episodes
1. Study Design — Where research begins
Every study starts with a question. How do we frame PICO/PECO? How do we choose outcomes? How do we protect against bias? The daughter asks her father about the very starting point of clinical research.
- A Story of Coffee Chat and Research Hypothesis
- Data Have Types: A Coffee-Chat Guide to R Functions for Common Outcomes
- Outcomes: The Bridge from Data Collection to Analysis
- A First Step into Survival and Competing Risks Analysis with R
- When Bias Creeps In: Selection, Information, and Confounding in Clinical Surveys
- A Story of Coffee Chat and Research Hypothesis Research often stumbles first not on data, but on how a question is framed.
- Data Have Types: A Coffee-Chat Guide to R Functions for Common Outcomes Numbers may look the same, yet describe different worlds—this pause starts from outcome types and the probability models behind them.
- Outcomes: The Bridge from Data Collection to Analysis Once an outcome is defined, much of the study is already decided—pausing over that choice through familiar oncology examples.
- A First Step into Survival and Competing Risks Analysis with R Before analysis begins, trouble often begins with how events are defined and coded—thinking in terms of time and events.
- When Bias Creeps In: Selection, Information, and Confounding in Clinical Surveys Distortion rarely comes from formulas alone—this is where bias quietly enters.
2. Glossary — The words of statistics
3. Frequentist Thinking — Practicing statistics
Using cancer clinical trials as an example, my father explains the frequentist perspective of “How do numbers behave under repeated sampling?”
- Reading a Paper over a Cup of Coffee
- P-Value Explanations That Seem Plausible at First Glance
- Beyond 0.05: Interpreting P-Values in a Clinical Trial
- Reading a Paper over a Cup of Coffee The unease felt while reading a paper often comes from language — pausing over statistical terms.
- P-Value Explanations That Seem Plausible at First Glance What would happen if the same study were repeated again and again — touching the idea of frequentist thinking.
- Beyond 0.05: Interpreting P-Values in a Clinical Trial A p-value is less a number than a shadow of design — watching where interpretation drifts.
4. Frequentist Experiments — Validation of statistics
The laws derived from the frequentist theory are not mere abstractions; they can be verified through simulation experiments. Even if you are unfamiliar with R, simply run the script to reproduce the results.
- R Demonstration of Bias in Kaplan-Meier Under Competing Risks A curve that looks convincing may still describe another world — checked through simulation.
- Understanding Confidence Intervals via Hypothetical Replications in R Where does “95%” come from, and where does it go — lingering over confidence intervals.
- Alpha, Beta, and Power: The Fundamental Probabilities Behind Sample Size Planning of sample size is also the planning of error of the study. Revisiting the foundations of frequentist design.
5. Effects and Time — How effects evolve
We unravel risk differences, risk ratios, hazard ratios, vaccine effectiveness, attributable fractions, and effects that change over time alongside survival curves and cumulative incidence curves. A multidimensional perspective on data quietly develops, paving the way for causal inference.
- [Silent Confusions Hidden in Percentages]
- [Who Is This Percentage About? Target Populations and Attributable Fractions]
- [Understanding Collapsibility of Effect Measures: Marginal vs Stratified]
- [When Odds Ratios Approximate Risk Ratios—and When They Fail]
- [From Risk and Rate to Survival and Hazard]
- [Distinguishing Time-Point, Time-Constant, and Time-Varying Effects: An R Example]
- Silent Confusions Hidden in Percentages Percentages feel intuitive, yet they often mislead. Rethinking how effects are communicated.
- Who Is This Percentage About—Target Populations and Attributable Fractions Every number speaks about someone — reconsidering the assumed population.
- Understanding Collapsibility of Effect Measures: Marginal vs Stratified When adjustment changes results, the reason is structural — revisiting stratification.
- When Odds Ratios Approximate Risk Ratios—and When They Fail Measures that look similar may behave differently — pausing over the odds ratio.
- From Risk and Rate to Survival and Hazard Once time enters the picture, effects change shape. An entry point to thinking about effects over time.
- Distinguishing Time-Point, Time-Constant, and Time-Varying Effects Are effects ever constant — considering time-varying perspectives.
6. Adjusting for Bias — The landscape of regression modeling
An episode where she learns from her father — from interpreting logistic regression results and creating tables to avoid pitfalls in its use. And then…
- [From Risk to Logistic Regression]
- [Logit: How a Transformation Shapes an Effect]
- [Where My Logistic Regression Went Wrong]
- [Why Logistic Regression Fails in Small Samples]
- [Understanding Confounding in Effect Measures: Marginal vs Stratified]
- [Volatility, Uncertainty, Complexity, and Ambiguity in Causal Inference]
- From Risk to Logistic Regression Translating risk into equations inevitably hides something — an entry into regression.
- Logit: How a Transformation Shapes an Effect Why this transformation and not another? A secret behind the choice of logit.
- Where My Logistic Regression Went Wrong When calculations are correct but conclusions feel wrong — pausing at interpretation.
- Why Logistic Regression Fails in Small Samples A model that breaks may still be teaching something — about small samples.
- Understanding Confounding in Effect Measures: Marginal vs Stratified When overall and subgroup results disagree — revealing a structure of confounding.
- Volatility, Uncertainty, Complexity, and Ambiguity in Causal Inference What exactly is the purpose of the DAG tool? A daughter’s response to her father’s teachings.
7. Truth — What defines it?
- [What Data Cannot Tell Us]
- [What Could Have Happened]
- [What Structures Structure]
- [What Is It That You Want to Know?]
8. Causal Inference — To find causality
Directed acyclic graphs (DAGs), common causes, colliders, mediators, backdoor criteria. We try to talk about the language of causality needed to think about “what if?”, incorporating the minimal necessary mathematical expressions concerning probability.
- [Three-Variable DAGs: The Smallest Building Blocks of Causal Structure]
- [A Subtle Distinction between Common Causes and Confounders]
- [DAGs and Conditional Distributions: Two Languages for the Same Structure]
- [A Circle, an Equation, and a Cylinder]
- [Backdoor Paths, Block, and d-Separation: A Clue for Adjusting for Bias]
- Three-Variable DAGs: The Smallest Building Blocks of Causal Structure The smallest diagrams that let us speak about causality.
- A Subtle Distinction between Common Causes and Confounders Similar words, different roles—tracing the boundary between common causes and confounding.
- DAGs and Conditional Distributions: Two Languages for the Same Structure Diagrams and equations tell the same story in different languages.
- A Circle, an Equation, and a Cylinder Where causal frameworks overlap and where they diverge — placing causal models side by side.
- Backdoor Paths, Block, and d-Separation What it really means to “adjust” — Understanding blocking as an operation.
9. Publish a Paper — Beyond the research
How to communicate research findings, creating charts and graphs, tips for revising, navigating peer review. A quiet lecture for researchers, exploring what lies beyond statistics.
- [A Subtle Distinction Between Editors and Reviewers]
- [Three Tips for Writing a Paper]
- [Communicating with Care]
- [A Morning Just Before Submission]
This site shares selected R-related posts via R-bloggers
