Spatial Aggregation and the Ecological Fallacy

Wakefield, Julie; Lyons, Hil

doi:10.1201/9781420072884-c30

Cited by 57 publications

(62 citation statements)

References 74 publications

(93 reference statements)

Supporting

Mentioning

Contrasting

Order By: Relevance

“…In particular, it also allows us to model both exposures and disease risk as spatially continuous processes, irrespective of the resolution at which incidence data have been recorded. This has the main benefit that assessment of the regression relationships is not affected by the bias induced by the ecological fallacy (Wakefield & Lyons, 2010), thus overcoming the limits of other approaches based on Markov structures (e.g. Held et al (2005)).…”

Section: Discussionmentioning

confidence: 99%

Modeling of spatio-temporal variation in plague incidence in Madagascar from 1980 to 2007

Giorgi

Kreppel

Diggle

et al. 2016

Spatial and Spatio-temporal Epidemiology

View full text Add to dashboard Cite

Plague is an infectious disease caused by the bacterium Yersinia pestis, which, during the fourteenth century, caused the deaths of an estimated 75 to 200 million people in Europe. Plague epidemics still occur in Africa, Asia and South America. Madagascar is today one of the most endemic countries, reporting nearly one third of the human cases worldwide from 2004 to 2009. The persistence of plague in Madagascar is associated with environmental and climatic conditions. In this paper we present a case study of the spatio-temporal analysis of plague incidence in Madagascar from 1980 to 2007. We study the relationship of plague with temperature and precipitation anomalies, and with elevation. A joint spatio-temporal analysis of the data proves to be computationally intractable. We therefore develop a spatio-temporal log-Gaussian Cox process model, but then carry out marginal temporal and spatial analyses. We also introduce a spatially discrete approximation for Gaussian processes, whose parameters retain a spatially continuous interpretation. We find evidence of a cumulative effect, over time, of temperature anomalies on plague incidence, and of a very high relative risk of plague occurrence for locations above 800 meters in elevation. Our approach provides a useful modelling framework to assess the relationship between exposures and plague risk, irrespective of the spatial resolution at which the latter has been recorded.

show abstract

Section: Discussionmentioning

confidence: 99%

Modeling of spatio-temporal variation in plague incidence in Madagascar from 1980 to 2007

Giorgi

Kreppel

Diggle

et al. 2016

Spatial and Spatio-temporal Epidemiology

View full text Add to dashboard Cite

show abstract

“…With the exception of the geostatsp package, the model in Equation 14 can not be fitted in the other packages reported in Table 1 other than by replacing individual-level explanatory variables d ij by their locationlevel averages. However, this would invalidate inferences on the regression coefficients β by introducing ecological bias (Wakefield and Lyons 2010).…”

Section: Conclusion and Future Developmentsmentioning

confidence: 99%

PrevMap: An R Package for Prevalence Mapping

Giorgi¹,

Diggle²

2017

J. Stat. Soft.

View full text Add to dashboard Cite

In this paper we introduce a new R package, PrevMap, for the analysis of spatially referenced prevalence data, including both classical maximum likelihood and Bayesian approaches to parameter estimation and plug-in or Bayesian prediction. More specifically, the new package implements fitting of geostatistical models for binomial data, based on two distinct approaches. The first approach uses a generalized linear mixed model with logistic link function, binomial error distribution and a Gaussian spatial process as a stochastic component in the linear predictor. A simpler, but approximate, alternative approach consists of fitting a linear Gaussian model to empirical-logit-transformed data. The package also includes implementations of convolution-based low-rank approximations to the Gaussian spatial process to enable computationally efficient analysis of large spatial datasets. We illustrate the use of the package through the analysis of Loa loa prevalence data from Cameroon and Nigeria. We illustrate the use of the low rank approximation using a simulated geostatistical dataset.

show abstract

“…Ecological bias is the difference between estimated associations on ecological-and individual-level data [83]. Data used in DM and ESR, both for the number of cases and for the covariates are found rarely at individual-level, mainly due to confidentiality reasons, and therefore the association found at the aggregated level might not be the same if we would have used individual-level data.…”

Section: 3 S Ta N Da R D I Z E D M O R B I D I T Y R At I Omentioning

confidence: 99%

“…Aggregated data is usually designated as areal data [5]. The objective of DM is not to estimate the associations between the cases and the covariates or to improve predictions, and therefore ecological bias is not a concern (for more details on the subject see Wakefield and Lyons [83]). …”

Section: 3 S Ta N Da R D I Z E D M O R B I D I T Y R At I Omentioning

confidence: 99%

A Gaussian random field model for similarity-based smoothing in Bayesian disease mapping

Baptista

Mendes

MacNab

et al. 2016

Stat Methods Med Res

View full text Add to dashboard Cite

Conditionally specified Gaussian Markov random field (GMRF) models with adjacency-based neighbourhood weight matrix, commonly known as neighbourhood-based GMRF models, have been the mainstream approach to spatial smoothing in Bayesian disease mapping. In the present paper, we propose a conditionally specified Gaussian random field (GRF) model with a similarity-based non-spatial weight matrix to facilitate non-spatial smoothing in Bayesian disease mapping. The model, named similarity-based GRF, is motivated for modelling disease mapping data in situations where the underlying small area relative risks and the associated determinant factors do not vary systematically in space, and the similarity is defined by “similarity” with respect to the associated disease determinant factors. The neighbourhood-based GMRF and the similarity-based GRF are compared and accessed via a simulation study and by two case studies, using new data on alcohol abuse in Portugal collected by the World Mental Health Survey Initiative and the well-known lip cancer data in Scotland. In the presence of disease data with no evidence of positive spatial correlation, the simulation study showed a consistent gain in efficiency from the similarity-based GRF, compared with the adjacency-based GMRF with the determinant risk factors as covariate. This new approach broadens the scope of the existing conditional autocorrelation models.

show abstract

Spatial Aggregation and the Ecological Fallacy

Cited by 57 publications

References 74 publications

Modeling of spatio-temporal variation in plague incidence in Madagascar from 1980 to 2007

Modeling of spatio-temporal variation in plague incidence in Madagascar from 1980 to 2007

PrevMap: An R Package for Prevalence Mapping

A Gaussian random field model for similarity-based smoothing in Bayesian disease mapping

Contact Info

Product

Resources

About