Creating Groups from Data. Cluster Analysis and Latent Class Analysis

About

Workshop Contents and Objectives

In this course, you will learn how to identify groups in data by working through the full analytical life cycle, from data preparation and variable selection through model estimation, validation, interpretation, and presentation of results, using both your own data (if available) and replicating published studies, in a transparent and reproducible manner using R.

You might, for example, want to distinguish types of web users based on variables that capture aspects of online activity (such as content preferences or time spent online). Or you may wish to study vaccine hesitancy by identifying groups of people who share similar concerns. In such settings, we often work with large datasets and many potentially relevant variables, so it is not feasible to detect groups of similar cases by browsing raw data or scanning tables.

To discover groups in data, we will use cluster analysis and latent class analysis. These techniques allow you to describe the phenomenon of interest in a structured way and to examine how group membership (such as user type) relates to other variables (such as gender, socioeconomic status, life satisfaction, or personality).

Cluster analysis is an unsupervised machine learning approach that applies various algorithms to identify similar cases (for example persons, organisations, schools, or countries) in your data. It groups similar cases into a specified number of clusters that are as distinct from one another as possible.

Latent class analysis, in contrast, starts from a probabilistic model that explains group membership. It works with the observed distribution of your data using a statistical model to estimate latent classes. You can include covariates, and the procedure provides goodness-of-fit measures that you can use to compare alternative model solutions.

The course introduces both cluster analysis and latent class analysis. We will spend two and a half days on cluster analysis (including hierarchical clustering, non-hierarchical clustering, k-means, and fuzzy clustering for continuous, categorical, and mixed variables) and two and a half days on latent class analysis (including latent profile analysis and longitudinal applications). We will also consider recent developments and applications, building on neural network-based approaches and density-based clustering.

You will work with data provided by the instructors, and you are encouraged to bring your own data if possible.

By the end of the course, you will understand the logic, advantages, and limitations of cluster analysis and latent class analysis, and you will be able to apply these techniques in R to your own research data.

Workshop design

Fifty-fifty mix of interactive lectures and hands-on exercise sessions. Exercises can be done individually or in groups. Daily feedback and possibilities for individual consulting.

Detailed lecture plan (daily schedule)

Day 1
Introduction to Cluster Analysis, Hierarchical Cluster Analysis

Day 2
Non-Hierarchical Clustering

Day 3
Fuzzy Clustering and Advanced Applications

Day 4
Introduction to Latent Class Analysis, Modelling Covariates

Day 5
Advanced Latent Class Analysis (Constrained, Multigroup, Longitudinal), Latent Profile Analysis.

Course feedback and Q&A.

Class materials

Slides
Exercise scripts, annotated solutions, additional illustrative scripts (all in R)
Set of exemplary papers
Selected readings by topic

**The Summer School cannot grant credits. We only deliver a Certificate of Participation, i.e. we certify your attendance.**

If you consider using Summer School workshops to obtain credits (ECTS), you will have to investigate at your home institution (contact the person/institute responsible for your degree) to find out whether they recognise the Summer School, how many credits can be earned from a workshop/course with roughly 35 hours of teaching, no graded work, and no exams.

Make sure to investigate this matter before registering if this is important to you.

Course leader

Robin Samuel is an Associate Professor at the University of Luxembourg (Department of Social Sciences) and Head of the Centre for Childhood and Youth Research. Hamid Bulut is a sociologist and researcher with a focus on computational social science. He holds a PhD in Sociology from the University of Luxembourg, complemented by a background in Business Administration.

Target group

graduate students, doctoral researchers, early career researchers, experienced researchers

Prerequisites

Participants should be familiar with univariate and bivariate statistics, including bivariate correlation and chi-square (for example in the context of crosstabs, also known as contingency tables). If you have not previously encountered these topics, this course is probably not appropriate for you. Ideally, you will also have some knowledge of OLS and logistic regression.

The course uses the statistical software R. R can be used to run cluster analyses and finite mixture models, such as latent class analysis and latent profile analysis. To benefit fully from the course, you should already be familiar with R (for example importing data, running basic commands, and inspecting output). Participants without prior R experience can still follow if they have worked with other syntax-based statistical software (such as Stata or SPSS), are comfortable learning new software, and are prepared to invest additional time during and between sessions. In all cases, you must be able to perform basic data management tasks in R or another package, such as recoding variables and handling missing values.

Here are some helpful materials for those who are new to R or feel they would benefit from a refresher: https://stats.idre.ucla.edu/r/

We will use the software R. R allows running cluster analyses and finite mixture models (e.g., latent class analysis and latent profile analysis). While familiarity with R would be useful, this is not strictly necessary if you have some knowledge of working with other statistical software packages using syntax (e.g., Stata or SPSS) and are willing to learn. You must be able to perform basic data management tasks in R or another software (e.g., recoding of variables, missing values, etc.).

Here are some helpful materials for those who are new to R or feel they would benefit from a refresher: https://stats.idre.ucla.edu/r/

Fee info