Statistical analysis

For the 2-day experiment, a generalized mixed linear model for a Poisson distributed outcome was used. Because the data exhibited over-dispersion, a negative binomial analysis was used. The fixed effects were month, treatment (Anthrone vs. ASB), and the interaction of month and treatment. Treatment is a repeated measure, and a random error term of village nested in month was used to provide an error term for the repeated measure. A heterogeneous compound symmetric covariance matrix was used to represent the correlated data structure. Model mean percent, standard error, and 95% CI of the difference between means as well as P-value for a comparison between treatments at each month is presented. Population density had Poisson distribution. Over-dispersion was evident; therefore, a generalized mixed linear model for a negative binomial distribution to analyze the data for each of the three trap types: CDC, Malaise, and PSC was used separately. The model included fixed effects for month (April–December), treatment (control and experimental: a repeated measure over months), and the interaction of month and treatment. A random error term of villages nested within treatment was used for the error term for treatment. A compound symmetric covariance matrix was used to represent the correlated data structure. Model means and standard errors as well as 95% confidence intervals (CIs) for mean differences are presented for the interaction. P-values are also presented for planned comparisons between treatments at each month. Human landing catches also had a Poisson distribution with over-dispersion; the same analysis plan was used as described above for the traps. The gonadal age and sporozoite infection rate data both had binomial distributions, therefore, a generalized linear mixed model was used to analyze these data. The same model as for the trap data described above was used for both. Model mean and standard error as well as 95% CIs for mean differences are presented for the interactions. P-values are also presented for planned comparisons between treatments at each month. A general linear model analysis for repeated measures was performed for the monthly EIR rates for indoor and outdoor data separately. The outcome was the EIR; predictors were group (control vs. treatment), month (April–December) and the interaction of group and month. A random effect of village nested within group was included to provide an error term for the repeated measures over month. A compound symmetric covariance matrix was used to represent the correlated structure of the data. Model means and standard errors were reported as well as P-values for comparisons between the groups at each month. The percent reduction in EIR due to the treatment is also included. The two-tailed alpha level was used to determine statistical significance. SAS 9.4 (SAS Institute Inc; Cary NC) was used for all analyses.