Take the 2-minute tour ×
Geographic Information Systems Stack Exchange is a question and answer site for cartographers, geographers and GIS professionals. It's 100% free, no registration required.

I am working on an epidemiological model with disease incidence rate as dependent variable and environmental factors as independent variable. Disease data is available at county scale whereas environmental data is in raster format that can be re-sampled as needed. Scale plays a major role when we talk about modeling as a model may be good at one scale and not so good at the other. For this reason I think conducting sensitivity analysis may be the right option. My question here is how to approach based on the data available for study. Any ideas, links to articles or journals is much appreciated.

share|improve this question
    
1. What software are you comfortable with? 2. Is there a time component in your data? –  radek Jul 19 '13 at 14:24

1 Answer 1

Since your disease data is aggregated at the county level, I'd start with exploratory analysis using this aggregation.

Take the polygon boundaries of your counties and create set of variables from zonal statistics tool (R, QGIS, ArcGIS and others can do that for you). Mean/median values would be probably a starting point but min/max values could also be useful.

In the next step, you would have to find a way to build a model linking disease incidence to explanatory variable. Simple descriptive stats and correlations and OLS are good way to start.

At next stage you might want to take into account spatial dependency in your data. Do some ESDA using GeoDa software, R, ArcGIS - smooth the rates for counties with small number of cases, look at global and local indicators of spatial autocorrelation.

If spatial dependency is an issue - next steps would require some serious thinking how to account for that. Geographically weighted regression, multilevel modelling or Bayesian methods (have a look at INLA package) will be your friends here.

share|improve this answer
    
@ Radek, thanks for your useful advice. I have already implemented most of these by taking zonal statistics (Mean) for the independent variables and have also run GWR on my model. As these statistics are at county level aggregation, I suppose the results will differ at other aggregation. What is you suggestion for model justification at different scales (sensitivity analysis) and how should I go about doing smoothing of disease rate. Much appreciated! –  Abhishek Kala Jul 19 '13 at 15:06
    
You could aggregate your data to larger areas using one of the regionalization tools and see how that affects your results. Dig into the GeoDa manual - since your analysis is based on counts I think Empirical Bayes might be a way to go. –  radek Jul 19 '13 at 16:33

Your Answer

 
discard

By posting your answer, you agree to the privacy policy and terms of service.

Not the answer you're looking for? Browse other questions tagged or ask your own question.