Count data regression models: An application to the major rice insect pest counts in the Terai region of West Bengal

In spite of India's phenomenal rice production of 118.43 million tonnes in 2019–20, the potential yield and the yield realised at the farmers’ fields are vastly different. Among the factors contributing towards this yield gap, the infestation of insect pests causes significant economic damage. Since these biotic menaces are largely weather-dependent, weather-based predictions of insect pests are often utilised to make economic decisions about insect pest management. Hence, an effort has been made in this study to comparatively assess the suitability of different count data regression models for weather-based prediction of three major rice insect pests (viz., gundhi bug, brown planthopper and green leafhopper) in the Terai region of West Bengal. As the input weather variables are related in a linear fashion, principal components have been obtained to be utilised subsequently in the regression analysis. Among the regression models considered, the recently developed modified Poisson quasi-Lindley regression model has empirically outperformed all of its counterparts in handling over-dispersion. However, the Poisson regression model has provided better result when no over-dispersion is evident. Outcomes emanated from the investigation have also revealed that the over-dispersion test plays a fairly good role in providing reliable guidance on the presence of over-dispersion. Hence, it is suggested that before adopting any weather-based count data regression model for predicting insect pest counts, one needs to check whether the count response variable is indeed over-dispersed.