Allowing for cliffs and slopes in the risk surface when modelling small-area spatial data
Lead Research Organisation:
University of Glasgow
Department Name: School of Mathematics & Statistics
Abstract
Abstracts are not currently available in GtR for all funded research. This is normally because the abstract was not required at the time of proposal submission, but may be because it included sensitive information such as personal details.
Organisations
Publications
Lee D
(2011)
An extended conditional autoregressive model for Bayesian disease mapping
in International Statistical Institute World Statistics Congress
Lee D
(2011)
A comparison of conditional autoregressive models used in Bayesian disease mapping.
in Spatial and spatio-temporal epidemiology
Lee D
(2013)
Locally Adaptive Spatial Smoothing Using Conditional Auto-Regressive Models
in Journal of the Royal Statistical Society Series C: Applied Statistics
Lee D
(2012)
Boundary detection in disease mapping studies.
in Biostatistics (Oxford, England)
Lee D
(2013)
CARBayes : An R Package for Bayesian Spatial Modeling with Conditional Autoregressive Priors
in Journal of Statistical Software
Mitchell R
(2014)
Is There Really a "Wrong Side of the Tracks"in Urban Areas and Does It Matter for Spatial Analysis?
in Annals of the Association of American Geographers
Description | The main findings of the research were the development of new parsimonious statistical models to estimate the localised nature of spatial correlation, which is able to determine when two adjacent areas have spatially correlated risks and when they do not (a so called risk boundary). This methodology extends the majority of existing research, which assumes a global level of spatial correlation, i.e. sharing a common border equals spatially correlated risks. In addition, a small amount of literature exists on tackling this problem, but the methods proposed suffer from problems of parsimony and have latterly been shown not to be effective. The methodology described in the grant proposal was developed first, and has been published in two high-ranking peer-reviewed statistics journals (Biostatistics and Journal of the Royal Statistical Society Series C). These papers, together with the associated conference presentations, collectively present the following findings: • The new statistical methodology can accurately identify the locations of "risk boundaries" in simulated data. • The identification of such boundaries leads to more accurate estimation of both the spatial pattern in the risk surface and the effects of covariate factors than are obtained using existing global smoothing models. • The risks of alcoholism, cancer and respiratory disease in Greater Glasgow show the existence of many "risk boundaries", again suggesting that traditional global smoothing models are inappropriate. In conjunction with the papers a software package called CARBayes was produced for the statistical software R, which allows other researchers or public health professionals to apply the methods we have developed to their own data. The software goes further than what was promised in the grant proposal, and can fit the models developed to continuous (Gaussian) and discrete (Poisson and binomial) data. This software is an add-on package for the statistical software R. A paper describing the software has been published in the Journal of Statistical Software. The statistical methods developed have led us to investigate the factors that might cause "social boundaries" to occur, i.e. what causes rich and poor communities to live side-by-side? One possible explanation is the role of physical barriers, such as railway lines or rivers, which prevent the two communities from mixing. We have a paper in the Annals of the Association of American Geographers on this work. As a result of this research project we have been successful in obtaining grant funding for a follow on project from EPSRC (EP/J017442/1), which will extend the research developed here into the spatio-temporal domain. This will allow us to determine whether the locations of "risk boundaries" have changed over time, for example due to shifting population demographics. |
Exploitation Route | The main impact of the project to date has been scientific, and aimed at the wider academic community. This has come in the form of two novel statistical modelling approaches, which extend existing methods by allowing for localised rather than global spatial smoothing of areal-unit health data. Two peer-reviewed publications have resulted from this methodological development. This impact has been ensured by the creation of the CARBayes software, which is freely available for others to use via the R platform. This software has a growing user base, and I get regular questions about its use from both statisticians and non-statisticians alike. |
Sectors | Education,Environment,Healthcare |
Description | This research project has developed new statistical methodology for modeling localized spatial autocorrelation, and as such its non-academic economic and societal impact was always likely to be limited, especially in comparison with its scientific impact. However, the project has had the following impact in this area. • The research has impacted upon the work of the Glasgow Centre for Population Health (GCPH), because it appears in their recent report in April 2013 entitled "Investigating the impact of the spatial distribution of deprivation on health outcomes". • Government agencies such as Health Protection Scotland (HPS), the Information Services Division (ISD) of the National Health Service in Scotland, and the Scottish Government have been informed of our research via a one-page executive summary/briefing note we created. • A blog entitled "What if neighbouring areas are very different" posted about our work on our Centre for Research on Environment, Society and Health (CRESH) website. |
Sector | Healthcare |
Impact Types | Societal |
Description | Carnegie-Caledonian PhD Scholarships |
Amount | £64,000 (GBP) |
Organisation | Carnegie Trust |
Sector | Charity/Non Profit |
Country | United Kingdom |
Start | 10/2011 |
End | 02/2015 |
Title | CARBayes |
Description | The software CARBayes is a package for the statistical software R, and can be downloaded from the CRAN website at https://cran.r-project.org The software fits Bayesian hierarchical models for spatial areal unit data, using Markov Chain Monte Carlo simulation. |
Type Of Technology | Software |
Year Produced | 2012 |
Open Source License? | Yes |
Impact | I receive regular correspondence from academics working in a variety of disciplines who use the software asking for guidance on how to use it. |
URL | https://cran.r-project.org/web/packages/CARBayes/index.html |
Description | A blog about the project |
Form Of Engagement Activity | A magazine, newsletter or online publication |
Part Of Official Scheme? | No |
Geographic Reach | International |
Primary Audience | Public/other audiences |
Results and Impact | We wrote a blog for the website of the Centre for Research on Environment, Society and Health about the project. This has widened the impact of the project for people interested in this area. |
Year(s) Of Engagement Activity | 2012 |
Description | An invited Applied Quantitative Methods Network lecture. |
Form Of Engagement Activity | A talk or presentation |
Part Of Official Scheme? | No |
Geographic Reach | Regional |
Primary Audience | Other academic audiences (collaborators, peers etc.) |
Results and Impact | An invited lecture at a one-day workshop hosted by the ESRC funded Applied Quantitative Methods Network (AQMeN). The audience were mainly social scientists and epidemiologists, and this gave me the chance to disseminate my research outside the statistics community. This talk made this research area more accessible to social scientists, and resulted in new collaborations between myself and colleagues such as Dr Mark Livingston. |
Year(s) Of Engagement Activity | 2012 |
Description | An invited seminar at the University of Bath |
Form Of Engagement Activity | A talk or presentation |
Part Of Official Scheme? | No |
Geographic Reach | National |
Primary Audience | Other academic audiences (collaborators, peers etc.) |
Results and Impact | I gave a research seminar at the University of Bath, which shared this research project with my peers in the fields of statistics and epidemiology. Prompted a thoughtful discussion afterwards amongst my peers, which highlighted the importance of the problem of boundary detection in spatial data. |
Year(s) Of Engagement Activity | 2011 |
Description | Briefing note |
Form Of Engagement Activity | A press release, press conference or response to a media enquiry/interview |
Part Of Official Scheme? | No |
Geographic Reach | National |
Primary Audience | Policymakers/politicians |
Results and Impact | We created a briefing note / executive summary about the project for Health Protection Scotland. This has resulted in dialogue between HPS and the project group. |
Year(s) Of Engagement Activity | 2012 |
Description | Talk at the ISI conference |
Form Of Engagement Activity | A talk or presentation |
Part Of Official Scheme? | No |
Geographic Reach | International |
Primary Audience | Other academic audiences (collaborators, peers etc.) |
Results and Impact | Around 40 people attended the conference talk, which disseminated the findings of the research project. None |
Year(s) Of Engagement Activity | 2011 |
Description | Talk at the RSS conference |
Form Of Engagement Activity | A talk or presentation |
Part Of Official Scheme? | No |
Geographic Reach | International |
Primary Audience | Other academic audiences (collaborators, peers etc.) |
Results and Impact | Around 30 people attended the conference talk, which disseminated the findings of the research project. None |
Year(s) Of Engagement Activity | 2012 |
Description | Talk at the Spatial statistics conference |
Form Of Engagement Activity | A talk or presentation |
Part Of Official Scheme? | No |
Geographic Reach | International |
Primary Audience | Other academic audiences (collaborators, peers etc.) |
Results and Impact | Around 50 people attended the conference talk, which disseminated the findings of the research project. None |
Year(s) Of Engagement Activity | 2011 |