Please note: This course will be taught online only. In person study is not available for this course.
Alexandru Cernat is an associate professor in the social statistics department at the University of Manchester. He has a PhD in survey methodology from the University of Essex and was a post-doc at the National Centre for Research Methods and the Cathie Marsh Institute. His research and teaching focus on: survey methodology, longitudinal data, measurement error, latent variable modelling, new forms of data and missing data. You can find out more about him and his work at: www.alexcernat.com
Longitudinal data are an essential tool for researchers as they can help answer questions about change in time, causal relationships and the timing of events. They come in many shapes, from traditional panel surveys to social media and sensor data. Because of their additional complexity, specialized statistical models are needed to analyse them.
In this course you will learn how to analyse longitudinal data using R. The course is developed to include statistical models from a number of different fields, giving students a comprehensive knowledge of models such as: multilevel models for change, latent growth models, cross-lagged models and survival models. The course is also hands, each topic being accompanied by real world applications using R and practical exercises. In addition to learning about statistical models the students will also learn how to prepare and visualize longitudinal data. They will also have the opportunity to discuss about their own research projects and get guidance on how they can use the methods covered in the course in their own work.
To gain competence in the concepts, designs and terms of longitudinal research;
To be able to apply a range of different methods for longitudinal data analysis;
To have a general understanding of how each method represents different kinds of longitudinal processes;
To be able to choose a design, a plausible model and an appropriate method of analysis for a range of research questions.
The course will be based on the free and open source software R. As such, a basic working knowledge of R is needed at the start of the course. Also, the course assumes a working knowledge of linear and logistic/probit regression.
Representative Background Reading (both books will be provided by ESS):
Singer, J., & Willett, J. (2003). Applied longitudinal data analysis: modeling change and event occurrence. Oxford University Press.
Newsom, J. T. (2015). Longitudinal Structural Equation Modeling: A Comprehensive Introduction. Routledge.
Background Knowledge Required:
Calculus = Elementary
Linear Regression = Moderate
OLS = Moderate
Maximum Likelihood = Elementary
R = Elementary
POTENTIAL ESS APPLICANTS ARE TO BE ADVISED THAT RECORDINGS WILL NOT BE MADE AVAILABLE FOR THIS COURSE.
In each day we will cover one overarching topic. The day will be divided into a lecture, a hands on practical using real data and a recap session in which we will go through the solution together. Students will be able to ask questions throughout the day.
Below you can find the topic and the reading for each day.
Day 1: Longitudinal data preparation
One of the most time consuming aspects of longitudinal data analysis is the data preparation phase. In this day we will discuss the main concepts of longitudinal data and then we will focus on preparing a large panel study for analysis. On the way you will learn how to import data, merge it, clean it and reshape it.
Chapters 2, 3 in Long, J. D. (2011). Longitudinal Data Analysis for the Behavioral Sciences Using R. Thousand Oaks, Calif: SAGE Publications, Inc.
Day 2: Longitudinal data visualization
Before analysing our data it is a good idea to explore it. This will enable us to find issues, such as coding errors, as well as inform future research questions and modelling decisions. In this day you will learn how to explore longitudinal data. We will focus especially on the use of visualization as a tool for understanding longitudinal data.
Chapter 4 in Long, J. D. (2011). Longitudinal Data Analysis for the Behavioral Sciences Using R. Thousand Oaks, Calif: SAGE Publications, Inc.
Day 3: Introduction to Structural Equation Modelling and auto-regressive models
Structural Equation Modelling (SEM) is a flexible framework in which a number of different longitudinal models can be estimated. In this day you will learn how to use the SEM framework to run simple longitudinal models. This will be a foundation on which to build more complex models in the next days.
Chapters 1, 4 in Newsom, J. T. (2015). Longitudinal Structural Equation Modeling: A Comprehensive Introduction. Routledge.
Day 4: Introduction to cross-lagged and longitudinal mediation
In this day we will use the SEM framework to estimate two longitudinal models. The cross-lagged model is used to understand the causal direction when we have two variables of interest that are strongly related. We will also look at how the popular mediation model can be used in a longitudinal context.
Chapter 5 in Newsom, J. T. (2015). Longitudinal Structural Equation Modeling: A Comprehensive Introduction. Routledge.
Day 5: Introduction to fixed effects and the multilevel for change
In this day we will discuss two of the most popular longitudinal models used in the social sciences: fixed and random effects. You will also learn how these models are related to the multilevel model for change and what are the strengths and limitations of the different approaches.
Chapters 3, 4 in Singer, J., & Willett, J. (2003). Applied longitudinal data analysis: modeling change and event occurrence. Oxford University Press.
Day 6: Advanced topics in multilevel model for change
In this day we will delve deeper in the multilevel model for change. We will explore how to include time varying predictors and how these could be re-coded to help with results interpretation. We will also discuss about non-linear change in time and about different strategies to estimate such models.
Chapters 5, 6 in Singer, J., & Willett, J. (2003). Applied longitudinal data analysis: modeling change and event occurrence. Oxford University Press.
Day 7: Introduction to Latent Growth Modelling
In this day we explore an alternative model for estimating change in time: the Latent Growth Model (LGM). While this model is similar to the multilevel model for change it uses the SEM framework. We will discuss advantages and disadvantages of this approach as well as the similarities to the multilevel model for change.
Chapter 7 in Newsom, J. T. (2015). Longitudinal Structural Equation Modeling: A Comprehensive Introduction. Routledge.
Day 8: Advanced topics in Latent Growth Modelling
In this day we will delve deeper in the LGM and discuss how to include time varying predictors in the model and how to also run the parallel LGM. We will also explore how to estimate non-linear change in time using LGM.
Chapter 8 in Newsom, J. T. (2015). Longitudinal Structural Equation Modeling: A Comprehensive Introduction. Routledge.
Day 9: Longitudinal equivalence and second order LGM
The SEM framework makes it possible to empirically investigate if concepts of interest are measured the same way across different points in time. This is an essential assumption made by all longitudinal models. In this day we will cover how we can test this assumption and how we can build second order LGM that can correct for measurement error.
Chapters 3, 5 in Little, T. D. (2013). Longitudinal Structural Equation Modeling. Guilford Press.
Day 10: Survival analysis for discrete and continuous time
In this day we will discuss models used to explain the occurrence and the timing of events. We will discuss both discrete survival models and continuous survival models (Cox models).
Chapters 9. 10, 11, 13, 14 in Singer, J., & Willett, J. (2003). Applied longitudinal data analysis: modeling change and event occurrence. Oxford University Press.