Home  >  Sites  >  Workshops and Workflows

Statistical Modelling

This pathway consists of:

  • Three workshops for researchers interested in statistical methods such as linear regression, ANOVA, ANCOVA, mixed models, logistic (binary) and Poisson (count) regression. Each one builds on the preceding workshop showing how all these analyses can be performed using the same easy to understand Generalised Linear Mixed Model (GLMM) framework and workflow, and how they can be used to analyse experimental designs such as Control vs Treatment, Randomised Control Trials (RCTs), Before After Control Impact (BACI) analysis, repeated measures, plus many more.
  • A fourth complementary workshop called Statistical Model Building which we recommend for those experienced with Linear Models or for those who have done at least the first two of our Linear Models workshops.
  • The material is organised around Statistical Workflows, applicable in any software, giving practical step-by-step instructions on how to do the analysis, including assumption testing, model interpretation, and presentation of results. Some workshops also have accompanying Software Workflows for R.

Linear Models 1: Linear Regression, ANOVA, ANCOVA and Repeated Measures (a Simple Mixed Model)

In this workshop we focus on practical data analysis by presenting statistical workflows applicable in any software for four of the most common univariate analyses: linear regression, ANOVA, ANCOVA, and repeated measures (a simple mixed model) – all assuming a normal (gaussian) residual. These workflows can be easily extended to more complex models. There is also an accompanying software workflow for R.​

Linear Models 2: Logistic and Poisson/Count Regression - An Introduction to Generalised Linear Models

​In this workshop we focus on practical data analysis applicable in any software for two of the more common GLMMs: Logistic regression for binary data (using a Binomial distribution); and Poisson/count regression for count data (using a Poisson distribution). The GLM framework is also described in detail. There is also an accompanying software workflow for R.​

Linear Models 3: How to build interpretable models and analyse data to extract insightful & impactful patterns, and craft an engaging research story

Statistical analysis is more than just building the best predictive model, it should also enable you to make impactful discoveries that expand our knowledge. Constructing engaging narratives about your research is also invaluable as you look to connect with your field, the community and funding bodies. To do this you need to build interpretable models, test hypotheses, uncover insightful & impactful patterns, and present results in insightful, intuitive and memorable ways. In this workshop we explore tips and tricks to make your research do just that. Topics covered will be:

  1. Building impactful real-world recommendations and guidelines – i) why we need to understand both stated and model derived importance, ii) how Quadrant Analysis uses both variable performance and importance to develop impactful real-world recommendations and guidelines.
  2. Reporting tricks that extract insightful & impactful patterns and craft engaging stories – i) establishing the importance of a predictor/risk factor, ii) confidence vs prediction intervals, iii) applying and correcting for multiple comparisons, iv) testing different hypothesis using different model parameterisations of the design matrix, v) interpreting categorical predictors - dummy vs effects coding and estimated marginal means, plus other reporting and interpretation tricks.
  3. Building interpretable models – it’s quite common for researchers to incorrectly use model parameters to establish variables ‘impact’ or ‘importance’ . We show how multi-collinearity prevents this interpretation, and how to assess and then fix it so parameters can be used to identify important predictor/risk factors and other insightful patterns.
  4. Mixed models – extend the Linear Model 1 intro to: i) better explain how mixed models work, ii) use them to test population wide hypotheses outside your sampled groups, iii) use a random slope (with examples of the patterns it can explain and hypotheses it can test).
  5. Using data visualisation to report complex nonlinear models graphically and aid pattern extraction

Statistical Model Building

In this workshop we will introduce you to the key aspects and strategies of statistical model building to help you answer your research question, and avoid common pitfalls, erroneous models and incorrect conclusions. Appropriate statistical model building will help you to gain knowledge, as opposed to simply getting the best prediction (although that can be a goal as well). We will focus on concepts such as variable selection, multi-collinearity, interactions, selecting a model building strategy, comparing models and evaluating models. In general, these concepts are useful for any statistical model building. This workshop will provide generalised linear regression model examples. The focus will be on practical application of concepts, so mathematical descriptions will be kept to a minimum.