The University of Arizona

Research Methods | CLIMAS

Research Methods

We used a spatial model, the Genetic Algorithm for Rule-set Production (GARP) to provide insight into the climatic and environmental characteristics of the habitats of mosquitoes that transmit WNV. The GARP model combines ecological niche theory and empirical field work, through a combination of artificial intelligence programming and statistical optimization, to inductively define a theoretical area where a species can sustain a population. For Ae. aegypti, this niche may be characterized by a variety of biophysical and societal variables. Different spatial and temporal patterns of these variables form the best predictive models for each mosquito species. We explored novel methods of spatially analyzing an ecological niche to further understand WNV transmission at different spatial and temporal extents. We derived a rule set describing mosquito presence, which was iteratively developed using a process of data sampling, rule selection, evaluation of the goodness of fit, preservation or rejection of a rule, and additional rule generation until there was minimal improvement in predictive accuracy.

Our dengue fever research used data on the presence or absence of Ae. aegypti at 68 sites in Tucson and Nogales, Arizona and Nogales, Sonora (Mexico) during the months of July, August, and September. We also used microclimatic variables and housing characteristics data.

We used multivariate logistic regression models (a variant of standard linear regression used when the dependent variable is a dichotomy, such as success/failure) to determine the relative importance of the independent variables to mosquito presence the week before mosquito collection and the four days of mosquito collection. We generated regression models for the months of July, August, and September. In addition to individually analyzing presence/absence of the mosquitoes in each month of each year, a Generalized Estimating Equation (GEE) was used to analyze presence/absence for each month across all years of data collection. A GEE is a covariance pattern model that directly models the correlation of repeated measures on each site.