Models and the Pillars of Research with Data
Disclaimer
Depending on who you ask, this talk is about:
- Epidemiology
- Data science / analytics
- Machine learning / Artificial intelligence
These are generally the same, but differ in the terminology used for the same concepts, certain convictions, and approaches.
Agenda
- Tests are Void in the Presence of Models
- What Makes a Model?
- Making a Model
- The Pillars of Data Research
- Models and Pillars
- Modelling in Practice
Tests are Void in the Presence of Models
What is a Test?
Statistical tests are tools made for hypothesis testing.
They yield the p-value, but do not say anything about direction or size.
Examples are:
- T-test
- Chi-square test
- Fisher’s exact test
What is a Model?
Statistical models are mathematical models that summarise the relationship between different variables
This relationship is quantified and can be used for a plethora of goals beyond hypothesis testing.
Statistical tests are just simple statistical models:
| Statistical test | Equivalent statistical model |
|---|---|
| One-sample t-test | Intercept-only linear regression |
| Wilcoxon signed-rank test | Ranked univariable linear regression |
| Two-sample t-test | Univariable linear regression |
| One-way ANOVA | Multivariable linear regression |
| Chi-square test | Univariable logistic regression w/ dichotomous independent variable |
Models > Tests
Models allow us to:
- Model more explicitly
- Model more flexibly
- Derive more information from our data
What Makes a Model?
Parametric & Non-parametric models
The End
Contact me: r.j.janse-5@umcutrecht.nl
More about me: rjjanse.github.io
These slides: rjjanse.github.io/talks/modelling
Image for title slide by Environmental Graphiti:
350 Species at Risk from Climate Change
© 2025 Environmental Graphiti® All rights reserved.