Analysis of Overdispersed Insect Count Data from an Avocado Plantation in Thika, Kenya

Abstract

Avocado (<i>Persea americana</i>) farming in East Africa has expanded since recent, contributing significantly toward economic growth and livelihood for small-scale farmers. However, insects attacking avocado fruits reduce fruit quality and size, causing massive losses. Previous studies have identified key avocado insect pests, their temporal population patterns and how landscape vegetation productivity influences their population dynamics. This research analyzed insect count data collected on <i>Bactrocera dorsalis</i> and <i>Ceratitis</i> spp. in an avocado plantation in Thika, Kenya over a successive period of time, as part of pest management. These data are characterized by overdispersion due to aggregation behaviour of the insects in their habitat and serial correlations since the count data were collected over a successive period of time. Analyzing these data becomes complicated because of overdispersion and the serial correlation in the data. In this study, we explored variants of generalized linear models (GLMs) with a sinusoidal component over time; and with and without timescale decomposition of covariates (weather variables). All GLM variants were fitted assuming the negative binomial distribution to account for overdispersion. Based on the Akaike information criterion (AIC), GLMs with decomposed covariates had lower AIC values than GLMs without decomposed covariates for both <i>B. dorsalis</i> and <i>Ceratitis</i> spp., and therefore GLMs with a sinusoidal component and decomposed covariates under negative binomial distribution were the best choice for these data. The contribution of the preceding weekly insect pest counts in all models was statistically significant. The study established that both abiotic and biotic factors drive insect pest infestation.