Computational protocol: Oviposition Site Selection by the Dengue Vector Aedes aegypti and Its Implications for Dengue Control

Similar protocols

Protocol publication

[…] Regression analyses were conducted using R version 2.8.1 . To check for spatial autocorrelation among containers surveyed in the same week as a potential confounder, we estimated Moran's I for egg counts using a Euclidean distance matrix with the APE package within R . Because no spatial structure was evident, subsequent analyses did not take spatial coordinates into account. We attempted to include the density of adult female Ae. aegypti as a predictor variable in our models, but collections were too sparse (mean = 0.14±0.52 SD females per house) for meaningful analyses. Instead, using a separate chi square test, we examined whether the presence of Ae. aegypti larvae was independent from capture of adult females during the abundance survey.To identify variables that best predicted whether or not female Ae. aegypti laid eggs in a container, a logistic regression model was fitted to our data (1 = container received eggs at least once during three days of observation, 0 = container received no eggs). Categories of Ae. aegypti larval abundance were further divided depending on whether larvae were retained during the survey or removed from containers on the first day. Collection period was included to control for time. The three measures of container size were collinear (circumference-capacity, Spearman's ρ = 0.86; circumference-water volume, Spearman's ρ = 0.65, capacity-water volume, Spearman's ρ = 0.85). Because the amount of space available for oviposition is determined by container circumference, we included circumference rather than capacity or water volume in our model. Larval abundance and estimated mean larval density also were collinear (Spearman's ρ = 0.92). Larval abundance was used because it provided a better model fit to the data. Starting with a saturated model including all variables listed in , we employed a log-likelihood test to eliminate, stepwise, the non-significant variable with the greatest χ2 p-value (2× log-likelihood of current model–2× log-likelihood of previous model ∼χ2, df = 1, p>0.10). If the final model included a variable with more than two levels, Tukey's multiple comparisons were applied using the MULTCOMP package to identify differences in level effects.Only containers receiving eggs were included in the analysis to identify variables influencing the number of eggs laid in containers. Negative binomial regression was performed using the MASS package . Our response variable was the mean number of eggs laid per container per day, rounded to the nearest integer. To more closely examine the association between egg abundance and container size, we included both container circumference and (circumference)2 as predictor variables in the model. As with the logistic regression model, containers were classified according to larval abundance and whether or not larvae were removed on the first day, and to collection period to control for time. Model selection was based on the log-likelihood test. To confirm that model assumptions were met, deviance residuals were plotted against: (1) fitted values, (2) each explanatory variable included in the model, (3) each explanatory variable eliminated from the model, (4) survey date, and (5) spatial coordinates . […]

Pipeline specifications

Software tools APE, multcomp
Applications Phylogenetics, GWAS
Organisms Aedes aegypti