The initial impact of COVID-19, social distancing and movement restrictions on crime in NSW

The Canberra Branch of the Statistical Society meeting on Tuesday 30 August was addressed by Dr Joanna Wang of the University of Technology, Sydney. Over 20 people joined the online seminar.

Joanna began by reviewing the literature, which included studies from the US, Sweden, UK and Canada. Her won work used data from NSW from 2017 – mid-2020, a timespan ensuring that the first COVID lockdown was represented. She focused on a handful of different crime types, noting that there is debate around whether the counts should be aggregated or analysed separately, and whether to expect disaggregated estimates to sum up to aggregated totals.

Joanna used a classic interrupted time series model, though I was curious to learn more about the ARIMA error structure that she was also able to impose. As always, there’s an R package for that (fable in this case) along with the use of AIC for model selection.

The main conclusions were that the change (mostly decline) in crime during the early days of COVID was more about the changing supply of opportunities than the changing supply of perpetrators. Joanna did a great job of presenting the model and results in a clear fashion, along with discussion of what it all meant.

Measuring Sex, Gender Identity and Sexual Orientation

The AGM of the Statistical Society of Australia took place on Tuesday 30 August and as is traditional at Branch AGMs, it was followed by a presentation. This particular talk by a trio of United States researchers attracted over 100 audience members. Diane Herz of the Social Research Centre introduced the work of the Centre and the day’s speakers.

Nancy Bates, Kellan Baker and Bianca Williams all played a role in the production of this important NIH sponsored report, available here. Each of them presented part of the work, from the background and data collection principles to the key recommendations.

Data collection principles encompassed values such as inclusiveness, precision, autonomy, parsimony and privacy. There’s potential for many strains between these. The recommendations included ones around whether sex or gender should be the variable collected “by default”, and options for multiple choice questions on sexual orientation including two-stage questions. The US also has First Nations whose voice has become more important in recent times, and the notion of “two-spirit” as a placeholder for nuanced First nations conceptualisations of sexual identity was introduced.

Finally, a call for more research! It makes perfect sense to statisticians that changes to the way data is collected, especially in long-established series, should be made on the basis of evidence about community understanding of the wide range of terminology in use around this fascinating topic.

Preliminary findings and implications from the August 2022 ANUpoll

Professor Nick Biddle of the Centre for Social Research and Methods presented this seminar online on Wednesday 24 August. I’ll be presenting in their seminar series soon so I was pleased to be able to join around 20 people online and another 10 at a watch party on campus.

Nick revealed some of the first results from the poll, focusing (as always!) on the effects of COVID. The particular outcomes of interest ranged from mortality and vaccination rates to long COVID. A new variable that caught my eye was stringency – a measure for 0 to 100 of the degree to which COVID restrictions such as stay-at-home, exercise limits and shop closures were imposed.

Nick used some interesting visuals involving differences in modelled probabilities to display the results of models. Of particular interest were determinants of vaccine uptake, including exposure to COVID itself.

The results Nick presented were for discussion not reporting so not actual figures in this report. However I’m sure we can look forward to publication of the final model results in the very near future!

What makes geospatial decisions different?

I’ve been on the seminar mailing list of the research School of Earth Sciences for quite a while, and this one stood out as being close to the statistical research and collaboration that I do. And so on Thursday 18 August I joined about 15 people online for this presentation by Dr Yongze Song of Curtin University.

And Songze did not disappoint in terms of statistical computational relevance. He is developing and using a variety of R packages all centred on geographical detection. Intriguing modifiers such as parameter-based, interactive, zone-base and robust flavour the different packages available. That said, traditional statistical methods such as stepwise regression, ridge regression and generalised additive models were also part of the toolkit. So were machine learning methods such as neural networks, random forests and support vector machines.

Songze based much of hi presentation around an application looking at the relative effectiveness of different sensor indicators in assessing infrastructure performance (in this case, road pavements).

I was excited to see my favourite stats package being extended in interesting directions for spatial applications. The style of data collected is somewhat different to the spatial data I am used to, and it would be a huge benefit if researchers in these disparate areas of application could come together to see where potential synergies lie.

The Matilda Effect – The History of Forgotten Women Scientists

It’s National Science Week, which means there’s usually some fun and different events celebrate science in all its manifestations.

One of those on Tuesday 16 August was this talk by Dr Louise Olsen-Kettle of Swinburne University of Technology. She is currently a Vice Chancellor’s Women in STEM fellow in mathematics. Her research is in building mathematical and computational models of damage to provide new understanding in forecasting risk and damage in a range of novel materials and resources. These discoveries are transferable across many industry sectors including composites, manufacturing, oil and gas, construction, and mining. 

The Matilda effect was named by Matilda Gage, who noticed that there is a bias away from acknowledging the achievements of women scientists, towards attributing their work to their male colleagues.

In this talk Louise introduced us to twenty women behind incredible scientific discoveries, inventions and innovations. From the abstract of the presentation, we read that “this included bacterial geneticist Esther Lederberg, who made amazing discoveries in genetics that won her first husband a Nobel Prize; astrophysicist Jocelyn Bell, who discovered the first radio pulsars in 1967, but was excluded from the Nobel prize awarded to her thesis supervisor Antony Hewish and astronomer Martin Ryle. A similar fate befell Rosalind Franklin, the chemist excluded from the Nobel prize awarded to her colleagues James Watson, Francis Crick, and Maurice Wilkins for the discovery of DNA, and Lise Meitner who led the research that ultimately discovered nuclear fission, however it was Meitner’s colleague Otto Hahn who received the accolades, a Nobel Prize in Chemistry and renown as the discoverer of nuclear fission. In the 1950s, physicist Chien-Shiung Wu devised a groundbreaking experiment to test the law of parity conservation, for which two male colleagues received a Nobel Prize.”

The women spanned the ages, from Theano who may have been the wife of Pythagoras, through 17th century scientists Maria Merian and Maria Kirch and 18th century scientists Sophie Germain and Mary Anning.

It wasn’t all bad news though, as Louise also introduced us to another twenty women who have been well rewarded for their scientific contributions, Notable in this list were Hypatia, Ada Lovelace, Florence Nightingale, Marie Curie, Grace Hopper and Jane Goodall. Current Antipodean heroines were not omitted, including Nalini Joshi, Lauren Gardner, and Siouxsie Wiles.

It was quite a whirlwind tour of names and contributions, and I was particularly pleased to see Florence Nightingale get a place in the honour roll. Amongst the Matildas, Emmy Noether also stood out to me as her picture graces the corridors of my office building.

The mathematics of bushfires

The Canberra Mathematics Association conference was on Saturday 13 August, and I was delighted to accept the invitation to speak. The theme was “Change – one of life’s constants?” so I decided to pin my talk to this theme by presenting on linear models. The Palmer penguins got a good workout, as did a variety of extensions to the simple linear regression models ranging from transforming the right and left hand sides to adding more error terms.

I was also pleased to hear Professor Jason Sharples of UNSW Canberra present this talk on his career’s work around bushfire modelling. The talk was in honour of the late Neville de Mestre, who not only worked on some of the first elliptic models of the mathematics of bushfires but also the mathematics of bodysurfing! Check out his scientific paper on this very Australian topic.

Jason drew not only on Neville’s work but also the Renaissance mathematician Christian Huygens, whose elliptical model served well but has been updated in a number of directions. Jason showed how the models need to include the wind that the fire clouds create needs to be included in modelling, to ensure that firefronts are modelled in the curved shapes that they follow. Log-linear models also come into play, which I was able to refer back to in my own presentation later that afternoon.

New Plumbing: Adding a pipe operator to Base R

I think this is the first Ihaka lecture I have attended, and Thursday 4 August was a great way to start. Online with over 30 people, and I reckon from the livestream that there were another 30 in person at the University of Auckland. All to hear Professor Luke Tierney of the University of Iowa speak on pipe operators in base R.

Who would have thought that such a topic could be so engrossing? Luke spoke about the famous pipe operator in the magrittr package, the proceeded to explain the issues that the base R developers had in deciding what features of piping were important enough to make it work in base R too.

One of the main issues to address was the way that pipes implicitly pass the output as the first argument of the new input. This is a problem for a function as fundamental as even lm(), because the data argument (generally the one you want to pass), is the second argument after the formula. Luke described several of the options that the developers canvassed before they settled on the one for release in 2022.

So for now, next time you are in R4.2.0, rty this: mtcars |> lm(mpg ~ disp, data = _)