March 28, 2023
The Canberra Branch of the Statistical Society invited the SPARSE team to present on our work at their March meeting. Alice, Sumon and Bernard presented in a triple act on the main methodology and results of the small area estimation project. We have the whole of Australia covered, now, as you can see from the slide behind Bernard in the three pictures of the hybrid talk below.
November 21, 2022
The APA conference is in Canberra later this week, and there is a Early Career Researcher symposium in the lead-up. Mu Li presented his work in progress on bivariate modelling of health outcomes using the Australian Early Childhood Development Census – here’s a photo of him discussing his poster, following on from a 15 minute oral presentation of the work earlier on the session.
November 16, 2022
It turns out we have managed to organise the SPARSE symposium on World GIS Day! The symposium is listed on the worldwide register of events marking the day, which will bring the statistical aspects of the project to a broader audience and strengthen the links between the mapping aspects of the project and other geographical information system activity.
September 8, 2022
An online seminar celebrated the achievements of the winners of the inaugural Venables Award for New Developers of Open Source Software for Data Analytics. Lydia Lucchesi and Sam Nelson were at the Canberra “watch party” to present on vizumap, including our use of the R package to illustrate childhood undernutrition in Bangladesh.
August 1, 2022
We’re delighted that Lydia Lucchesi and Sam Nelson chose to feature SPARSE in their interview with the Australian Data Research Commons. We’re using their software, vizumap, to illustrate our key research findings on smoking prevalence for disaggregated regions across Australia.
June 28, 2022
It’s like a five-yearly Christmas Day for statisticians and demographers, when the ABS releases the latest Census data. There’s been some conversation in the media pondering the relevance of the Census, but many researchers (including SPARSE!) are using Census data on a daily basis and are delighted to have this five-yearly anchor point for our research.
June 27, 2022
Congratulations to SPARSE AI Dr Petra Kuhnert whose mapping software vizumap has won the inaugural Venables Award for New Developers of Open Source Software for Data Analytics. We’re using vizumap to display the results of the smoking prevalence models we’re developing, and are delighted to have Petra on our team.
June 23, 2022
SPARSE team members A/Prof Alice Richardson, Dr Sumonkanti Das and Dr Bernard Baffour undertook a triple act at the NCEPH seminar series under the title “Mapping the prevalence of smoking with increased precision in sparsely sampled regions of Australia”. Alice introduced the research and placed it in the context of declining smoking rates with huge variation across disaggregated regions of Australia. Sumon described the data and modelling process followed by Bernard summing up the implications of the modelling for policy directions. As always the NCEPH audience (about two-thirds online, the rest in person) had thought-provoking questions about both the detail of the modelling and how best to convey these results to decision-makers.
May 31, 2022
The School of Demography offed a place at short notice in their seminar series, and SPARSE demographers Dr Sumonkanti Das and Dr Bernard Baffour stepped up to take the challenge. With the Small Area Estimation conference not long finished, Sumon spoke on “Estimation of daily smoking prevalence for disaggregated statistical areas in Australia” and revealed some of our newest maps, combining estimates with uncertainty right down to SA3 level across Australia.
May 23 – 27, 2022
The Small Area Estimation conference in Maryland, USA, offered both in-person and online attendance which was a huge bonus for the SPARSE team. Team members Dr Sumonkanti Das and Mu Li both presented talks online, as did one of our QUT supporters, Jamie Hogg; and Sumon’s co-authors Dr Jan van der Brakel and Dr Harm Jan Boonstra.
Sumon spoke on “Estimation of daily smoking prevalence for disaggregated statistical areas in Australia” and Mu spoke on “Constructing a socio-economic index at disaggregated statistical area level in Australia for spatio-temporal modelling”.
Professor Andrew Gelman was one of the keynote speakers, his topic was “Small area estimation models: when do they fail?”. It was a pleasure to be able to watch a presentation from such a well-known statistical thinker and modeller. A fuller report is available on Alice’s blog.
May 17, 2022
Emeritus Professor Steve Haslett is a Visiting Fellow at RSFAS at ANU, among other roles, including that of AI on the SPARSE project. Steve presented the School of Demography seminar in person on Tuesday 17 May, and the Canberra SPARSE team joined over a dozen people in person, and another half-dozen online, for the talk.
Steve has a wealth of experience working with international organisations such as the World Food program, as well as over 20 national statistical organisations. His expertise has been called upon by this wide range of countries to implement small area estimation of measures such as nutrition, health and poverty. The importance of the small area estimates is that they allow funding agencies to be more certain of the distribution of the outcome of interest, so that programs to combat it can be deployed where they are needed and the success or otherwise of the program measured with reasonable accuracy.
Steve described the small area estimation model and showed examples of its use in settings ranging across Asian countries. He also discussed whether or not small area estimates are useful for ranking because of the standard errors associated with them. But perhaps more importantly he drew our attention to issues of data quality that can derail even the best small area models. Questions around how to define poverty, where to draw the line, and even questions as simple as what is this house actually built of and is it fit for purpose, were all part of Steve’s insights. He also spoke of the importance of local context and on-the-ground experience, including food, geography and even the presence or absence of mosquitoes as important factors in the statistical models.
April 20, 2022
The SPARSE team has a great collaboration going with the Australian Cancer Atlas team. SPARSE AI Susanna Cramb supervises a PhD candidate, Jamie Hogg, who presented his thesis confirmation seminar. The photo below appeared on Twitter, and the abstract of the talk is given below. Well done Jamie on a great presentation!
Understanding geographic differences in the prevalence of cancer risk factors are most valuable for the design of community-level interventions for cancer prevention. In Australia, cancer risk factor prevalence data is not available at the smallest geographic scale and is not available for very remote areas. In this project, we will develop small area estimation methods that target the data sparsity setting of Australia to produce SA2-level small area estimates of the prevalence of cancer risk factors. Following this work, we will explore the spatial variation of cancer risk factors and area-level associations between specific risk factors. Finally, we will provide an alternative and broader summary of the current prevalence of cancer risk factors through the creation of a set of area-level cancer risk indices for several cancer types. These indices will be beneficial for policymakers wishing to project future cancer risk in Australia.
December 5 – 10, 2021
The International Population Congress took place virtually in 2021.
We’re delighted to report that our poster “Trends in chronic undernutrition in Bangladesh for small domains using Bayesian hierarchical time series modelling” was awarded “Best poster” on Day 3 of the Congress. This is a fantastic affirmation of this work which uses many of the same techniques that we’ll be applying to measure and track smoking prevalence in Australia over time.
December 9, 2021
Sumon Das, Research Fellow working on the SPARSE project, presented the seminar titled “Multilevel time series modelling of antenatal care coverage in Bangladesh at disaggregated administrative levels” to the Research School of Finance Actuarial Studies and Statistics at the ANU.
In this study Sumon and colleagues applied multilevel time series (MTS) models to estimate trends in time series based on Demographic and Health Surveys from 1994 to 2014. The models are expressed in a Bayesian framework with Normal priors, though Fay-Herriot style estimates are also used to calculate direct estimates that feed into the models too. The trend estimates at national level behave in the expected manner, with a gradual decline in women receiving no antenatal care, and a gradual rise in women receiving the preferred four antenatal care visits. At division (seven of those) and district (64 of those) level though, the trends provided by the MTS models are much more variable including unexpected rises and falls over the time period. The MTS models are able to smooth many of these anomalies, providing policy makers with better quality information on which to base their interventions and programs in the maternal healthcare space. The work will be re-submitted to the Journal of Survey Methodology in the next few weeks.
October 15, 2021
Our project team members are active in a wide variety of statistical applications, and we are happy to draw attention to it when their work appears in public forums. Professor Steve Haslett co-authored a Conversation piece on How a random sampling regime could help detect COVID and highlight infection hotspots. As COVID restrictions start to ease in the countries in which SPARSE team members live, we are keen to hear about ideas that would help health authorities and the general public to navigate “COVID normal” living.
September 20 – 24, 2021
Sumonkanti Das, Research Fellow on the SPARSE project, attended the virtual Small Area Estimation conference for 2021, hosted in Naples, Italy, with the theme “Big Data for Small Areas”. He presented on multilevel time series modeling of mobility trends in the Netherlands for small domains, by accounting for the impact of survey redesigns and COVID-19 on mobility trend, in an invited session on “Time series methods for Small Area Estimation” organized by Professor Jan van den Brakel. There were some talks highly linked with the SPARSE project. In an invited session on issues and opportunities from record linkage and data integration in SAE, Professor Gauri Datta presented a generalized Fay-Herriot model by accounting measurement errors in covariates. This generalized model is required when contextual variables extracted from a sample of Census data are utilized in development of area-level SAE model. Professor Ray Chambers organized a plenary session on the honour of Dr. Hukum Chandra, where Nikos Tzavidis and Nicola Salvati presented two important contributions of Hukum on outlier robust SAE and spatially nonstationary model for SAE. One of Hukum’s students (Dr. Saurav Guha) presented a R package called “NSAE” dedicated to spatially nonstationary SAE model. Some talks were on recently developed R packages dedicated to SAE methods, such as “mind” for multivariate model based inference for domains and “msae” for multivariate FH models for SAE. We would like to take some aspects of the SPARSE project to this conference next year.
August 24, 2021
Mu Li, PhD student on the SPARSE project, presented his thesis proposal review titled “Bivariate outcome small area estimation with spatio-temporal effects: the case of smoking prevalence in Australia” to the School of Demography.
Mu provided a detailed and comprehensive overview of the state of the art of small area estimation, including M-quantile regression and copula methods. His first year of PhD study has been dominated by coursework but Mu also spoke about a couple of side project involving estimation of US smoking prevalence rates, and developing a time-varying SEIFA index which will have the potential to feed directly into the application of the models he will develop in his thesis.
The models which Mu has in his sights are clever combinations of multivariate M-quantile regression and Bayes regression, with various combinations of spatial and temporal effects. I’m very excited about the prospects for this project as they will not only contribute to methodology in the small area estimation field, but will also contribute to the quality of estimates developed for SPARSE and the associated mapping of smoking prevalence.
August 10, 2021
Sumon Das, Research Fellow working on the SPARSE project, presented a seminar titled “Multilevel time series modelling of antenatal care coverage in Bangladesh at disaggregated administrative levels” to the School of Demography.
In this study Sumon and colleagues applied multilevel time series (MTS) models to estimate trends in time series based on Demographic and Health Surveys from 1994 to 2014. The models are expressed in a Bayesian framework with Normal priors, though Fay-Herriot style estimates are also used to calculate direct estimates that feed into the models too. The trend estimates at notional level behave in the expected manner, with a gradual decline in women receiving no antenatal care, and a gradual rise in women receiving the preferred four antenatal care visits. At division (seven of those) and district (64 of those) level though, the trends provided by the MTS models are much more variable including unexpected rises and falls over the time period. The MTS models are able to smooth many of these anomalies, providing policy makers with better quality information on which to base their interventions and programs in the maternal healthcare space. The work has been submitted to the Journal of Survey Methodology
August 10, 2021
August 10 is Census Day in Australia. The SPARSE team relies on Census data from 2011 and 2016 to produce the small area estimates of smoking prevalence that form the core of the project deliverables. We really encourage everyone in Australia on Census night 2021 to complete their online form. The data from this year will start to make an impact on our models from mid-2022, and we need everyone to respond fully in order to do our best work. Thanks in advance, “every stat tells a story”!
July 5-9, 2021
The SPARSE team submitted a poster to the Australian and New Zealand Statistics Conference. The conference was notionally on the Gold Coast, but all the talks and events were online from 5 – 9 July 2021. Because of timelines, the poster describes the aims of the project and the work the team plans to do. Next year the conference will be in Darwin and we hope to not only be able to report results from the SPARSE project but also present them in person.
May 18, 2021
School of Demography seminars. The School asks grant recipients to present a 30-minute talk on their project near the beginning of the work. Bernard Baffour presented on SPARSE. He introduced the project and talked about two supporting projects that are in progress while the application for Datalab access progresses. First, he spoke about the mapping of small areas estimates based on the Australian Early Development Index (work led by Mu Li and published in June 2020. Then he spoke about mapping chronic under-nutrition in Bangladesh across 22 years, 64 districts and yearly ages from 0 to 5 (work led by Sumon Das).
April 25, 2021
We were very saddened to hear over the Anzac weekend of the death from COVID-19 of Dr Hukum Chandra, Principal Scientist at the Indian Agricultural Statistics Research Institute in New Delhi, India. He completed his PhD in Social Statistics at the University of Southampton and his post-doctoral fellowship at the University of Wollongong, Australia. Dr Chandra has made outstanding research contributions to the discipline of statistics in general and survey sampling, in particular, for which he has been recognised globally. His research interests are in survey sampling design and analysis, small area estimation, bootstrap methods, statistical modelling and data analysis, and statistical methodological development in the area of agricultural statistics. He has received a number of awards for his excellent research contribution, such as the National Award in Statistics by the Indian Government’s Ministry of Statistics and Programme Implementation, the Cochran-Hansen Award by the International Association of Survey Statisticians for his PhD thesis, and the Young Researcher/Student Award of the American Statistical Association. He is a council member of the International Association of Survey Statisticians and an elected member of the International Statistical Institute (ISI).
Hukum visited Canberra in December 2019 just after the awarding of this grant, funded by a Cross-College Research Grant from the Research School of Social Sciences at ANU. We were looking forward to welcoming him again to a SPARSE symposium in mid-2022. Our thoughts are with his wife and children.
February 17-19, 2021
The Spatial and Temporal Statistics Symposium took place online, organised by students and staff at the University of Wollongong. Speakers included SPARSE team members Susanna Cramb and Petra Kuhnert. Susanna spoke on “TIPS (Ten Incorrect Pithy Statements) for Disease Mapping” and Petra spoke on “Reimagining science, the hybrid connection”.
September 17, 2020
School of Demography Work-in-Progress seminars. Mu Li presented his current PhD work on the SEIFA index. Contact Bernard Baffour for details (email email@example.com).
February 6, 2020
Alice Richardson attended the symposium “Advances in Statistical Methodology” to celebrate the career of Professor Ray Chambers. The one-day event took place at the University of Wollongong. Attendees included a number of key researchers in small area estimation, including Professor Nikos Tzavidis.
December 3 – 4, 2019
Bernard Baffour and Alice Richardson hosted a short visit to ANU by Dr Hukum Chandra. The visit was funded by a Research School of Social Sciences Cross-College Collaboration grant. Hukum presented a School of Demography seminar titled “Small area estimation of prevalence of diarrhoea among under-five children in Bangladesh by combining health survey and census data”. He also participated in a drop-in session with quantitative researchers in the Research School of Population Health, and visited the National Office of the Australian Bureau of Statistics in Canberra. Hukum, Alice and Bernard met with a range of interested groups within ABS including Ian Rayson, SPARSE Associate Investigator from the ABS Health Statistics area.