Data Analysis using Count Models

Description:

This is a two-day hands-on training workshop led by Dr Les Humphreys, Lancaster University, Dr Paul Norris and Dr Rebecca Pillinger, University of Edinburgh.

Course Content:
This course will explore methods for analysing aggregate crime data with a particular focus on analysing change over time. The value of these approaches in understanding levels of, and changes incrime is illustrated using examples of US recorded crime data in the lectures, and recorded crime data from Scotland in the practical sessions. This type of data is not normally distributed; it is count data and so takes discrete non-negative values which are Poisson or Negative Binomially distributed. Hence it is not appropriate to fit linear models to this type of outcome, as we will show. Instead, Generalised Linear Models (GLMs) should be used. These provide a method to estimate regression models where the distribution of the outcome variable belongs to a certain family of non-Normal distributions. This includes the Poisson distribution, on which we focus, and the Negative Binomial distribution, which we also discuss.

We will explore the basic principles of Poisson log-linear models and rate models. In addition, the course will cover the basics of changepoint models and spline models. Changepoint models allow us to detect whether and when changes in crime trends have taken place, and spline models allow us to estimate smooth regression curves. Whilst the substantive application is to crime, and thus the course is most relevant to students and researchers in this area, the techniques presented apply to any repeated measures of count data, and thus the course may also be of interest to those in other substantive areas who are engaging in research that explores change over time in social processes.

Course objectives:

• To show why GLMs are needed when working with count outcomes and provide a thorough introduction to using them
• To teach participants the principles underlying change point models and spline models and how these approaches differ from other methods of analysing repeated data
• To illustrate the application of these approaches using specific examples of recorded crime in Scotland and domestic violence in England and Wales
• To give participants practical hands on experience of fitting GLMs, changepoint models and spline models

Learning outcomes:

By the end of the workshop participants will:

• Understand the theory and methods behind GLMs, changepoint models, and spline models
• Be able to apply these techniques to suitable data making relevant assumptions
• Be confident in accurately interpreting the results of this type of analysis

Cost:

Various - Dependent upon position/job title

Website and registration:

https://www.aqmen.ac.uk/events/Jul2016/CountModels

Region:

Scotland

Keywords:

Data Quality and Data Management , Quantitative Data Handling and Data Analysis

Related publications and presentations:

Data Quality and Data Management
Quantitative Data Handling and Data Analysis

Back to archive...