TY - JOUR
T1 - Bayesian areal interpolation, estimation, and smoothing
T2 - An inferential approach for geographic information systems
AU - Mugglin, A. S.
AU - Carlin, B. P.
AU - Zhu, L.
AU - Conlon, E.
PY - 1999/8
Y1 - 1999/8
N2 - Geographic information systems (GISs) offer a powerful tool to geographers, foresters, statisticians, public health officials, and other users of spatially referenced regional data sets. However, as useful as they are for data display and trend detection, they typically feature little ability for statistical inference, leaving the user in doubt as to the significance of the various patterns and 'hot spots' identified. Unfortunately, classical statistical methods are often ill suited for this complex inferential task, dealing as it does with data which are multivariate, multilevel, misaligned, and often nonrandomly missing. In this paper we describe a Bayesian approach to this inference problem which simultaneously allows interpolation of missing values, estimation of the effect of relevant covariates, and spatial smoothing of underlying causal patterns. Implemented via Markov-chain Monte Carlo (MCMC) computational methods, the approach automatically produces both point and interval estimates which account for all sources of uncertainty in the data. After describing the approach in the context of a simple, idealized example, we illustrate it with a data set on leukemia rates and potential geographic risk factors in Tompkins County, New York, summarizing our results with numerous maps created by using the popular GIS Arc/INFO.
AB - Geographic information systems (GISs) offer a powerful tool to geographers, foresters, statisticians, public health officials, and other users of spatially referenced regional data sets. However, as useful as they are for data display and trend detection, they typically feature little ability for statistical inference, leaving the user in doubt as to the significance of the various patterns and 'hot spots' identified. Unfortunately, classical statistical methods are often ill suited for this complex inferential task, dealing as it does with data which are multivariate, multilevel, misaligned, and often nonrandomly missing. In this paper we describe a Bayesian approach to this inference problem which simultaneously allows interpolation of missing values, estimation of the effect of relevant covariates, and spatial smoothing of underlying causal patterns. Implemented via Markov-chain Monte Carlo (MCMC) computational methods, the approach automatically produces both point and interval estimates which account for all sources of uncertainty in the data. After describing the approach in the context of a simple, idealized example, we illustrate it with a data set on leukemia rates and potential geographic risk factors in Tompkins County, New York, summarizing our results with numerous maps created by using the popular GIS Arc/INFO.
UR - http://www.scopus.com/inward/record.url?scp=0032831192&partnerID=8YFLogxK
UR - http://www.scopus.com/inward/citedby.url?scp=0032831192&partnerID=8YFLogxK
U2 - 10.1068/a311337
DO - 10.1068/a311337
M3 - Article
AN - SCOPUS:0032831192
SN - 0308-518X
VL - 31
SP - 1337
EP - 1352
JO - Environment and Planning A
JF - Environment and Planning A
IS - 8
ER -