On 24 February Catherine d’Este from the Research School of Public Health gave this talk to kick off the year of talks at the Canberra Branch of the Statistical Society. She gave a really clear presentation, drawing on a number of interesting public health research questions dealt with in her group. I learnt that broken stick models are still current, and have been joined by bent cables in the ongoing attempts to model step changes in response to an intervention. I also learnt about step-wedge experiment designs, which also have a place in the staged introduction of changes in ABS-style surveys.
The guide was released in October 2014 as part of an OLT funded project “Developing a shared understanding of assessment criteria and standards for undergraduate mathematics”. Statistics has not been ignored in the project, and the guide includes a handy rubric for assessment of your typical first year statistics project.
The American Statistical Association has released a white paper on this meaty topic, and I really enjoyed reading it. The bits I highlighted for future use were this definition of statistics: “the science of learning from data, and of measuring, controlling and communicating uncertainty” and this request for the future: “it is essential that these data scientists possess a statistical perspective, which emphasizes the quality of the data, the power of sampling, the characteristics of the sample, the validity of generalization, and the balance of humans and machines.”
That’s the kind of graduate I’d like to see from my University – roll on the first semester of 2015 when we start the process again!
How much of published laboratory science should we believe? This was John’s subtitle for the talk he gave at ANU on 11 December. He started with delightful quote from the “Journal of Irreproducible Results” about the height of hats worn by witches, priests and chefs. Just like Terry Speed spoke about earlier this year, john emphasised the importance of estimating batch effects. I plan to keep an eye also on the Many Labs project, reproducing 13 classic psychological studies; Centreforopenscience.org, replicating Cancer biology studies; and validation.scienceexchange.com who will arrange for replication of your study for you. john also highlighted new guidelines for reporting research such as ARRIVE and REMARK. I’d like to find out more about the Harvard Institute for Quantitative Social Science too.
John will be returning to New Zealand soon – we will miss easy access to his R programming skills, and will have to keep in touch through his thoughtful and sometimes provocative ANZSTAT postings.
Jenny Chesters and Anne Daly managed to attract nearly a dozen people to this seminar on 10 December – Christmas may be coming but maybe the release of the 2014 NAPLAN report today inspired those still at work to come out and hear the talk.
They did have individual child-level NAPLAN data with the approval of DET, and fitted a nice hierarchical linear model in Stata. The claim was made that because this was population data (or at least 98% of the population as claimed in the newspapers today) then no confidence intervals needed to be presented, just the parameter estimates. Those estimates mostly looked quite significant in terms of percentage increase or decrease in scores though. Jenny did point out that if you’re desperate for a standard deviation of scores within a school, you could do worse than take the range ( which is apparently given on the MySchool website) and divide by 6. So long as you thought the NAPLAN score were normally distributed of course!
Part III of this book is on “Perspectives on the field and profession” and it’s been quite w hike since I posted on Parts I and II. This book has continued to be an absorbing read. Authors, chapter titles and a one-sentence reaction to each one follow.
Stephen Fienberg. Statistics in service to the nation. Four nice take-home messages after discussion of log-linear models.
Iain Johnstone. Where are the majors? Very short and puzzling.
Peter Hall. We live in exciting times. Recounts his rather charmed career and incredibly mathematical and theoretical nature of his research, though I’m sure he would beg to differ!
Rafael Irizarry. The bright future of applied statistics. Some good inspirational stuff for first lectures here.
Nilinjan Chatterjee. The road travelled: from statistician to statistical scientist. Pertinent comments about other researchers grabbing discoveries but statisticians carefully characterising results.
Xihong Lin. A journey into statistical genetics and genomics.
Mary Thompson. Reflections on women in statistics in Canada. Very Canadian focus, historical overview.
Nancy Reid. “The whole women thing”. Loved the call to get nominating, and to not be so hard on each other.
Louise Ryan. Reflections on diversity. Extremely valuable reflections on the experience of women in science in Australia even touching on overseas travel.
Professor Sonya Marshall-Gradisnik of Griffith University at the Gold Coast presented this John Curtin seminar on Friday 28 November. Her centre has attracted a variety of grants and has a strong structure to support research. She talked about a number of current projects, including a mobile clinic for moderate-severe cases of CFS in South-East Queensland, which has expanded into an app called CliniHelp. She presented the results of several published paper, dominated by bar charts with confidence integral sprouting out the top – not the most engaging of visuals but highly used in the medical research literature. I learnt a new phrase – cell lysis – and went away with some good ideas about the work-family balancing act from this highly successful female academic.
The weather in Nova Scotia is starting to turn bad, so Chris Field is back in sunny Canberra! His talk on Thursday 27 November was more of a tutorial on his thinking about this modelling problem in microbial counts. The particular issues to be grappled with included dealing with a multinomial distribution with over 1000 categories, and the fact that there is often only one replicate. We all leant a new word – mesocosm – and went away pondering possible alternative approaches ranging from WGCNA to principal components.
Anthony Morphett from the University of Melbourne gave this ontributed talk at EVIMS2. Deb King and Robyn Pierce are also involved in this project. He uses Geogebra and the list of examples included convergence of a sequence, differentiability of a function and power of a statistical test. I was pleased that he had picked upon the “abiding image” concept of Chris Wild.
Peter Eades from the University of Sydney spoke on this topic at a dual-badged seminar between EVIMS2 and the CSIT department at ANU.
A graph consists of nodes and edges, a rather specific definition which threw me form the beginning. But Peter was able to start with a triangle, with Minard’s graph as an example of communication; John Snows’s graph as an example of discovery; and pretty pictures of Internet connectivity as an example of “gee whiz” that Peter says is a rather discredited way to advance knowledge. Now it’s all very well to talk about removing edge crossings for readability, but there are no axes to the graphs Peter discusses, and no requirement to show the point (1,2) one above (1,1). onetheless after some surprising observations and a nostalgic theorem, stress calculations, Jaccard similarity and multidimensional scaling all got a look in as the talk concluded.