• What this module is about
• How the module fits together
• What the assessment asks you to produce
• What the assessment criteria reward
• What a good analysis report contains
• How to choose a workable dataset and question

This module is about using advanced regression models to answer research questions with data.

The practical goal is that you can produce reproducible analysis reports that show:

• why a model is appropriate
• how the model was fitted
• what the model estimates mean
• how well the model answers the research question

The module builds from linear regression towards more flexible regression models.

We use one broad modelling framework to understand:

• linear models for continuous outcomes
• logistic regression for binary outcomes
• count models for frequency outcomes
• multilevel models for grouped or repeated data
• one method that goes beyond the taught content

Four short analysis reports, each using a dataset of your own choice:

1. Binary logistic regression
2. Count model: Poisson, negative binomial, or zero-inflated
3. Multilevel regression model
4. A method beyond the class content

Each analysis should be no more than 4 pages (excluding references).

Deadline: Friday 4 December 2026, 2 pm via Dropbox.

You will need to work on these reports throughout the term, not just at the end.

Address some non-trivial question of some psychological relevance using your chosen data.

Psychology covers a wide range of topics and sub-fields, many of which overlap with other disciplines.

Consider the dataset described here, published in:

Bertrand, M., & Mullainathan, S. (2004). Are Emily and Greg more employable than Lakisha and Jamal? A field experiment on labor market discrimination. American Economic Review, 94, 991-1013.

The article was published in an economics journal, but it addresses implicit racial bias and is clearly related to psychology.

For each analysis, you should:

• state the research question and its psychological relevance
• describe the dataset, source, key variables, and sample size
• include useful descriptive statistics and visualisation
• describe the statistical model formally
• report fitted model results, including parameter estimates
• report model predictions
• compare plausible alternative models
• assess quality of the model

The formative assessment is one short linear-regression report.

Deadline: Friday 23 October 2026, 2 pm.

This assessment is supported during the workshops. Across the first three weeks, you will have time to work on:

• choosing a dataset and question
• understanding the structure of a reproducible report
• fitting and interpreting a model
• showing predictions
• comparing plausible models

From the NOW learning room, open formative-assessment-example.pdf.

Do not worry about how the report was created yet. For now, identify:

• where the research question is stated
• where the data are described
• where the table and figure are created
• where the model is presented
• where predictions are plotted
• where model comparison appears
• what parts you would need to replace for your own report

This is a good example of report structure, not a perfect report.

It is deliberately fairly simple so that we can see the main parts clearly. A marker might say that it could be stronger by adding:

• clearer motivation from psychological theory or previous research
• a more developed justification for the chosen predictors
• a richer model comparison linked more directly to hypotheses
• more detailed interpretation of the size and practical meaning of effects
• a stronger discussion of limitations and what the model cannot show

For a first-class report, the structure needs to be combined with more independent thinking and a more convincing argument.

A clear analysis report usually contains:

1. Research question
2. Dataset description
3. Descriptive statistics and plot
4. Model description
5. Model results
6. Predictions
7. Model comparison
8. Short conclusion

The rest of the session, and your homework, is dedicated to identifying a topic and dataset for the formative assessment.

There are two steps:

1. identify a topic and possible research question
2. search for and evaluate a dataset that could answer that question

We will do Step 1 first, discuss ideas as a group, then move to Step 2.

Finding a dataset is easier if you start with an area you already care about.

A good starting point is your project lab topic, or a topic from another psychology module.

Write down:

• one broad topic, such as sleep, anxiety, attention, memory, language, wellbeing, or ageing
• one behaviour, judgement, score, or response that could be measured
• one factor that might explain differences in that outcome

From the NOW learning room, open part-1-dataset-planning.html and use Exercise 1 as a guide.

Work in small groups and help each other. Do not search for datasets yet.

Keep in mind that the topic should connect to psychology, but do not worry about writing a full justification at this stage.

Answer the following questions (for yourself):

• one broad topic area
• one possible outcome
• whether that outcome could be continuous
• two or three possible predictors
• one possible research question

Each group should be ready to share one idea.

When listening to other groups, ask:

• Is the topic psychologically relevant?
• Is the possible outcome clear?
• Could the outcome be continuous?
• Do the predictors make sense for the question?
• Can the question be made simpler?

For the formative assessment, look for a dataset that can support a clear regression question.

A useful dataset usually has:

• a psychological or behavioural topic
• one outcome variable you can explain
• at least one meaningful predictor
• enough observations for a simple model
• documentation that explains the variables
• a file format you can open in R, such as CSV, Excel, SPSS, or R data

A continuous variable is numeric and has meaningful distances between values.

For a simple linear model, look for a rectangular table.

In this example, one row is one participant. The continuous outcome could be reaction time. Predictors could include age, sex, or smoking status.

First question for any dataset: what does one row represent?

Useful starting points include:

• Journal of Open Psychology Data: https://openpsychologydata.metajnl.com/articles
• OSF Search: https://osf.io/search/
• Rdatasets: https://vincentarelbundock.github.io/Rdatasets/datasets.html
• Kaggle Datasets: https://www.kaggle.com/datasets
• OpenPsychometrics: https://openpsychometrics.org/_rawdata/
• ICPSR: https://www.icpsr.umich.edu/sites/ICPSR/find-data
• Figshare: https://figshare.com/search
• Zenodo: https://zenodo.org/search
• OpenNeuro for neuroimaging data: https://openneuro.org/search

Search in two steps.

First, search for the topic and data source:

• sleep questionnaire open data
• memory reaction time dataset OSF
• anxiety wellbeing csv codebook
• attention experiment trial-level data

Then check whether the dataset has a usable outcome and predictors.

Do not choose a dataset just because the topic sounds interesting.

ChatGPT, Copilot, and similar tools can help generate search terms and identify possible repositories.

Useful prompts:

• “Suggest open psychology datasets about sleep with continuous outcome variables.”
• “What search terms could I use to find open data on attention and reaction time?”
• “For this dataset description, what could be a continuous outcome and predictor?”

Always verify suggestions yourself. Check the original source, documentation, licence, variables, and file format.

Kaggle is popular with students and can be useful, especially for CSV files with previews and column metadata.

Treat it as a starting point, not automatic approval.

A Kaggle dataset is only suitable if you can:

• identify the original source or documentation
• understand the column names and values
• find a usable outcome and predictors
• explain a psychological or behavioural research question
• check that reuse is allowed

Before choosing a dataset, check:

• file format is usable
• variables are documented
• one outcome variable is clearly identifiable
• predictors are meaningful and not too numerous
• there are enough observations for the planned model
• cleaning needs are manageable within the assessment
• reuse is allowed and ethically appropriate

Use your topic from Exercise 1 and the part-1-dataset-planning.html document to save notes.

By the end, identify:

• three search terms
• one online source you checked
• one candidate dataset
• where the documentation or codebook is
• one possible continuous outcome variable
• possible predictor variables
• whether the data could be analysed with a normal linear model
• one risk or difficulty with the dataset

For each candidate dataset, check:

• Does the research question match the data?
• Is the outcome variable clear and continuous?
• Could the outcome be analysed with a normal linear model?
• Is the source documented well enough?
• What would make the analysis simpler?

Use the remaining time to get help with these checks or to find a better dataset.

Useful books for this module:

• Andrews, Doing Data Science in R (Andrews, 2021): https://www.mjandrews.org/book/ddsr/
• Gelman, Hill, and Vehtari, Regression and Other Stories (Gelman et al., 2020): https://avehtari.github.io/ROS-Examples/