Modelling Strategies

Duration : 3.0 day(s)

:: Course Summary

This workshop provides participants with an array of efficient strategies for the building of robust and optimal regression models in a variety of situations. Is is relatively simple to carry out regression modelling with software packages nowadays. This workshop reviews pitfalls to avoid, good practices for data preparation, efficient tools for model validation, classical as well as more recent methods to use adjustments whenever necessary.

:: Learning Objectives

Upon completion of this workshop, participants will be able to:

Identify and avoid the most common errors in modeling;

Format, prepare and validate their data to make sure that they can run a regression analysis with their software packages;

Assess which is the most appropriate regression technique to use to solve a given problem;

Fit a regression analysis with their statistical software;

Produce insightful diagnostics and graphical summaries to assess model quality;

Measure the performance of a regression model;

Interpret and draw conclusions from the results of a regression;

Optimize a regression model by rectifying the problems identified and by determining the most appropriate relationship between the response and the explanatory variables.

:: Target Audience

This applied workshop is intended for non-statisticians who need to carry out data modelling - scientists, analysts, lab technicians, research assistants, engineers, graduate students, etc. It is also intended for statisticians interested in attending a practical workshop on data modelling. Because of its focus on good modelling practices, it is also relevant for newbies in statistics as for people who already have a working knowledge of regression techniques.

:: Prerequisite

Participants must have a good working knowledge of fundamental tools in statistics or they must have attended the training session Fundamental Tools in Statistics.

:: Topics Covered

General principle of data modelling
Different types of models
Choice of a type of model given the nature of the response variable
Choice of a type of model given the relationship between the response and the predictors
Data preparation for modelling
Measuring model performance
Diagnotic tools for models
Univariate vs. multivariate models
Specific models with multivariate models
Simple linear regression models
Multiple linear regression models
Handling redundancy in the explanatory variables
Variable selection
Models for categorical variables
Logistic regression
Specific tools to assess model performance in logistic regression: ROC curve
The notion of odds ratio
Alternatives to logistic regression
Non-linear regression
General considerations on reporting regression results
Use of models for prediction purposes

:: Course Content

This workshop is articulated around three sessions covering linear regression, regression models for categorical data and non-linear regression. It brings to the fore similarities and differences between these methods by focusing on the correct usage of each of them.

Among the covered topics, particular attention is paid to data preparation, missing data handling, choice of an appropriate modelling technique, the difference between explanatory and predictive models, handling redundancy in the predictors, variable selection and uncertainty measures on the predictions generated with models.

Moreover, an outlook of more recent modelling methods will be presented.

Finally, a important portion of the time is devoted to hands-on applications during which participants will have the opportunity to use their statistical software to build models using their own data sets that they are invited to bring along (a variety of datasets will be provided to participants who do not have the chance to use their own).