26-30 June 2017
This is a general course in data analysis using generalized linear models. It is designed to provide a relatively complete course in data analysis for post-graduate students. Analyses for many different types data are included; OLS, logistic, Poisson, proportional-odds and multinomial logit models, enabling a wide range of data to be modelled. Graphical displays are extensively used, making the task of interpretation much simpler.
A general approach is used which deals with data (coding and manipulation), the formulation of research hypotheses, the analysis process and the interpretation of results. Participants will also learn about the use of contrast coding for categorical variables, interpreting and visualising interactions, regression diagnostics and data transformation and issues related to multicollinearity and variable selection.
The software package R is used in conjunction with the R-commander and the R-studio. These packages provide a simple yet powerful system for data analysis. No previous experience of using R is required for this course, nor is any previous experience of coding or using other statistical packages.
This course provides a number of practical sessions where participants are encouraged to analyse a variety of data and produce their own analyses. Analyses may be conducted on the networked computers provided, or participants may use their own computers; the initial sessions cover setting up the software on lap-tops (all operating systems are allowed).
To introduce a theoretically consistent system of analysis that can be used to analyse a wide variety of data and research designs. Practical sessions will enable participants to analyse examples of all techniques using R and the R-commander.
Afternoon – Introduction: A system of analysis
Morning – Data coding, manipulation and management; defining models: representing research questions
Afternoon – Analysis: An introduction to generalized linear models; Interpretation: using effect displays
Morning – Modelling continuous data; Contrast coding: dealing with categories explanatory variables
Afternoon – Modelling count data; Including and interpreting interactions.
Morning – Modelling categories (using logit models); Modelling ordered categorial variables (proportional odds models)
Afternoon – Modelling unordered categorical variables (multinomial logit models); Exercises modelling categorical variables
Morning – Model diagnostics and data transformations (Box-Cox and Box-Tidwell); Variable selection (strategies for dealing with collinearity using limited variable models and multimodel presentations)
The course will be presented by Dr Graeme Hutcheson.
Graeme Hutcheson has worked in a number of social science disciplines for over 20 years and has written numerous books and academic papers dealing specifically with research methodology and statistics. His current interest is with the application and promotion of a unified system of analysis that applies to a wide range of research problems and can be learned within the time-frame available to a typical postgraduate student. He is currently working at Manchester University and also runs the Manchester R group, which promotes the use of ‘R’, the statistical analysis system.
Prior or recommended knowledge/reading/skills
There are no pre-requisites for this course as instruction is provided for all techniques. However, it will be of most use to those who are interested in modelling social science datasets (survey and quasi-experimental) and applying graphics to interpret these.
Agresti, A. (1996). An Introduction to Categorical Data Analysis. Wiley.
Fox, J. and Weisberg, S. (2011). An R companion to Applied Regression (second edition). Sage Publications
Harrell, F. E. (2001). Regression modelling strategies. Springer.
Hutcheson, G. & Sofroniou, N. (1999). The multivariate social scientist. Sage Publications.
Hutcheson, G. & Moutinho, L. (2008). Statistical modelling for management. Sage Publications.