The overall specific aim is to develop novel methods for the modeling and analysis of small area health data where different aggregation (AG) levels provide data support. In particular, we are interested in the ability of different AG levels to contribute to understanding unobserved effects at other levels.
The aims are: The development of multiple aggregation level methods to provide insight into aggregation effects at lower and higher levels. To develop novel methods to evaluate the effects of aggregation of spatial units on the behavior of small area relative disease risk models. The approach will examine the use of estimates from lower aggregation levels in aggregation estimators. A practical example of this would be using county level estimates to aggregate to estimates in public health districts (higher level health administration units). In addition we plan to examine the use of simex-like aggregational methods to provide insight into lower aggregation levels. To explore and evaluate models for missingness where outcomes are available in only some subareas. The development of methods for the spatial and spatio-temporal analysis of survey data, with individuals sampled in areas. Focus is on the investigation of missingness as a result of an incomplete sampling design, for example areas for which we want to make inference but which are not sampled, and missingness as a result of non-response, for example individuals that to do not answer specific questions. We want to investigate how to make inference at the aggregated level (the area) and examine the sensitivity to missingness in the spatial and spatio- temporal model components. The extension of the small area aggregation models in SA1 whereby biases in aggregation is considered with missingness will be considered. To examine the development of disaggregational effects in univariate and multivariate models. The assessment of latent effect models that allow disaggregation of inferential units. The assumption of space as effect modifiers is essentially the focus of this aim. We assume that 'natural'subsets of the study region have associated with them special models which form a diasaggregation of the overall regional model. A simple example of this is where a model allows for geographically adaptive regression so a covariate can have different effects in different areas. These effects could be continuous or discrete. When discrete they lead to unique spatial grouping/clustering of regression models.
We aim to examine both single disease and multiple disease examples of this phenomenon. The development of flexible software and the evaluation of models. In this aim we intend to develop a set of software programs that can be used to analyze aggregational problems described above. In addition we intend to develop and implement a range of model evaluation procedures which will allow the better comparison of methods proposed and their predictive capabilities.
The overall specific aim is to develop novel methods for the modeling and analysis of small area health data where different aggregation (AG) levels provide data support. In particular, we are interested in the ability of different AG levels to contribute o understanding unobserved effects at other levels. The proposed project has a potentially major impact in the field of public health (PH) for a variety of reasons. First, it focusses on readily available data held at different aggregation levels, the use of which can be the focus of public health studies and practice. Second, missingness in sample surveys is also a major practical concern for policy makers and we aim to address spatial aspects of such missing data. The problems of scale effects in disease risk analysis have not been addressed before in practical terms and we hope to provide good solutions to such problems that will be readily available. Finally, we will provide software to help in dealing with the range of problems outlined above, and this software will be user friendly and accessible to PH professionals and researchers.