Abstract
The impacts and limitations of personal protection measures against exposure to vectors of malaria and other mosquito-borne pathogens depend on behavioural interactions between humans and mosquitoes. Therefore, understanding where and when they overlap in time and space is critical. Commonly used approaches for calculating behaviour-adjusted estimates of human exposure distribution deliberately use soft classification of where and when people spend their time, to yield nuanced and representative distributions of mean exposure to mosquito bites across entire human populations or population groups. However, these weighted averages rely on aggregating individual-level data to obtain mean human population distributions across the relevant behavioural classes for each time increment, so they cannot be used to test for variation between individuals. Also, these summary outcomes are quite complex functions of the disaggregated data, so they do not match the standard binomial or count distributions to which routine off-the-shelf statistical tools may be confidently applied. Fortunately, the proportions of exposure to mosquito bites that occur while indoors or asleep can also be estimated in a simple binomial fashion, based on hard classification of human location over a given time increment, as being either completely indoors or completely outdoors. This simplified binomial approach allows convenient analysis with standard off-the-shelf logistic regression tools, to statistically assess variations between individual humans, human population subsets or vector species. Such simplified binomial estimates of behavioural interactions between humans and mosquitoes should be more widely used for estimating confidence intervals around means of these indicators, comparing different vector populations and human population groups, and assessing the influence of individual behaviour on exposure patterns and infection risk. Also, standard sample size estimation techniques may be readily used to estimate necessary minimum experimental scales and data collection targets for field studies recording these indicators as key outcomes. Sample size calculations for field studies should allow for natural geographic variation and seasonality, taking advantage of rolling cross-sectional designs to survey and re-survey large numbers of separate study locations in a logistically feasible manner.