Statistics and data science seminars

Meta-analysis is a powerful statistical technique for combining results across studies identified in a systematic review. There are useful frameworks and guidelines for carrying out and reporting meta-analyses, such as those provided by the Cochrane Collaboration.

This introductory talk will discuss some of the statistical challenges that arise once the studies for meta-analysis have been identified and the appropriate data need to be extracted. This stage in the meta-analysis is not always as straightforward as it first appears. This seminar will provide examples of common problems in deriving the data needed for the analysis and illustrate some solutions.

Linear models are the foundation for most classical methods in statistics; they are simple, powerful and readily interpretable. However, the linearity assumption does not always hold, and it can be useful to know how to relax this assumption. You can come to the wrong conclusion if you have "missed a squiggle" without knowing it! We will explore generalised additive models as a means of testing the linearity assumption and adapting the model using patterns in the data themselves. We will discuss how you can get the most out of such models, and the trade-offs involved in moving away from the well-worn path of linear models.

Data with a hierarchical structure is very common in many fields. Terms used to describe this kind of data include multi-level data, longitudinal data, split-plot designs or repeated measures designs.  These terms arose from different contexts, but they share a common feature. Observations are not independent, and there is structure to the dependence which can be incorporated into a statistical model.

Linear mixed-effects models ("mixed models") are a commonly used statistical method for multi-level or longitudinal data. Mixed models are versatile, capable of accounting for complex hierarchical structures, missing outcome data, and outcome variables with different distributions.

It is generally true that the more hypothesis tests you perform, the greater the probability that one of those tests is statistically significant by chance, the so-called multiple testing problem. At the Statistical Consulting Centre, we regularly receive enquiries regarding this issue. Reviewers often ask for adjustment for multiple comparisons. But how important is it? What can you do about it?

This seminar will outline some principles to consider, along with practical advice.

Researchers have access to a wide range of statistical software, but it can sometimes be challenging to decide which is the best choice to suit particular needs. This seminar will showcase some popular statistical software - R, SPSS and Minitab. We will discuss the considerations that can guide the choice of software based on individual preferences, needs and longer-term goals.

'You can’t fix by analysis what you bungled by design.' (Light, Singer and Willett, 1990) Good study design is one of the foundations of meaningful empirical research. In this seminar, we discuss the principles of good experimental design. These principles are important for anyone contemplating running an experiment. Associate Professor Graham Hepworth will share his vast experience in experimental design in around 40 years of consulting with researchers in agriculture, medicine and allied sciences, and many other disciplines. He will highlight examples of the best and worst in experimental design.

You’re a researcher at the University of Melbourne. Or a member of an analytics team in the finance industry. Or a market researcher. Do you need help with statistics or data science? Or something else again? In this seminar, I discuss these terms and how you can navigate your way through the sometimes bewildering array of concepts and activities that aim to support or collaborate with anyone pursuing quantitative inquiry.