led by Professor Paul Lambert, University of Stirling
In this session we introduce and reflect upon the role of multilevel models in applied social research, and provide practical training materials which demonstrate ways of running multilevel models in survey data analysis scenarios.
Multilevel models are popular in the social sciences as statistical analytical devices which can be useful in a variety of scenarios where data has a complex or ‘clustered’ structure. The most popular formats for running multilevel models are outlined in session materials and illustrated in multiple software environments (using Stata, SPSS and R). There remain plenty of situations, however, when the added value of using a multilevel model is ambiguous, and there are different views on the best strategies to use activities such as specifying, estimating, and interpreting suitable models. As well as giving introductory accounts, lecture and workshop materials also provide critical reflections on the place of multilevel models as statistical analytical procedures in applied social research, and describe and explore enduring debates about these methods.
The session activities will comprise approximately 3 hours’ worth of lecture-based materials and approximately 2 hours of practical workshop activities. Lectures are likely to comprise four talks each in sessions of around 45 minutes, and two practical workshops, each around 60 minutes. Lectures will introduce and reflect upon multilevel models in social research, with opportunities for questions and answers and discussion. Workshops will involve computer-based exercises which open existing datasets and run multilevel models on them, guided by software example files that are provided to participants.
The session is suitable for people who have at least some previous statistical analysis training and experience, and at a minimum are used to reading descriptive statistical results (tables and graphs), and have seen at least some examples of popular types of regression model (such as multiple regression models or logistic regression models). The session ought to be helpful for participants who have only a little statistical background, as it will use an introductory style to describe the features of multilevel models, and improve participants’ confidence in this approach. The session should also be useful to people who already have a more extensive statistical methodology background, as materials also provide critical reflections on strengths and limitations of the approach and explore selected advanced issues.