Individual exercises to help test your understanding and knowledge of key areas of text. Complete them to see your strengths and weaknesses.

The links will open in a new window.

Extra Resources

This additional resource accompanying the ‘Quantitative Modelling in Human Geography’ chapter is intended to provide you with hands-on experience of using multilevel models to explore the variability in body mass index (BMI) according to factors that relate to individuals, households and neighbourhoods (postcode sectors)[1]. This structure in the variability in BMI is plausible in that one can easily imagine determinants of BMI at the individual (e.g. predisposition to obesity and the knowledge, awareness and capacity to eat healthily or exercise), household (e.g. shared household diets) and neighbourhood (e.g. local food availability) levels. The analysis is based on a real large scale social survey, the Health Survey for England (2001‒4) (University of Manchester 2011).

The data and the practical workbook are available at:


The materials contain three practical exercise of which practicals 2 and 3 focus, in particular, on the use of multilevel models to analyse BMI decomposing the variability in BMI to levels of the individual, the household and the neighbourhood. The specific aims of the practicals are given below.

Practical 1: this practical explores the relationship between Body Mass Index (BMI) and various explanatory variables using the Stata statistical package. It provides context to later practicals by examining the relationship between weight and height in the data and implications for standard measures of Body Mass Index (BMI). Finally, it demonstrates how to convert Stata data files to MLwIN data files. The practical should be seen as an introduction to the data using Stata, the following practicals use MLwIN to exploit the multilevel structure of the data.

Practical 2: this practical models BMI using a set of increasingly complicated multilevel models from a variance components model to a random intercept and random slopes model. The analysis is performed using MLwiN (http://www.bristol.ac.uk/cmm/software/mlwin/), which is free to download for UK academics and students. The practical includes detailed notes on interpretation of results and the logic behind model development.

Practical 3: This practical extends the analysis of practical 2 to model BMI as a binary variable (overweight or not) through a multilevel logistic regression model.


University of Manchester. Cathie Marsh Centre for Census and Survey Research. ESDS Government. (2011). Health Survey for England, 2003‒2005: Multilevel Modelling Teaching Dataset. [data collection]. UK Data Service. SN: 6765, http://dx.doi.org/10.5255/UKDA-SN-6765-1.


[1] The practicals also include time (2001, 2001, 2003 and 2004) a fourth level of analysis within the model. However, as time contributes little explanatory power to the models, the practicals quickly drop this level of analysis.