# What are the most important aspects of modelling?

Module 4 of the Research Data Science course is an introduction to modelling. The aim is to give the students an overview of the basic principles of modelling, providing them with a rudimentary framework that they can apply to future modelling problems.

The data we will use is the [European quality of life survey](https://beta.ukdataservice.ac.uk/datacatalogue/studies/study?id=7316&type=Data%20catalogue), using the research question`What factors (material, occupational and/or psychosocial) predict self-reported health?` to focus our efforts.

We have one day, comprised on a taught session and a hands-on session. We will explore modelling using simple regression models, but we want to teach principles that are broader than regression and that could be applied to any problem where you attempt to mathematically extract regularities in noisy data (i.e. model your data).

Obviously this could be a full course by itself, so we want to whittle down what to teach to important and general principles.

**What are the most important aspects of modelling?**

Camila & Callum

---
(We've listed things that we think are important and general. Disagreements, additions, thoughts, and pithy rewordings are appreciated :grin:)

- Overfitting: fitting is easy, prediction is hard.
- Regularization
- Bias-variance tradeoff

- Models are always wrong. Model evaluation is about understanding why your model is wrong and whether the level of incorrectness is acceptable. :sleuth_or_spy:
- Modelling is finding patterns within uncertainty
- Probability distributions :game_die:
- Measurement matters: noise/uncertainty begins here
- How to quantify/report uncertainty.
- Weighted averages
- Models only know about the world you build for them
- Feature extraction and importance
- Training
- Validation

- Models explicitly test hypotheses.
- Prediction/simulation
- Counterfactual predictions (perhaps not)
- Baseline models

### References: