We used quasi-generalized linear models with aPoisson error (qGLMs) to analyse the relationshipbetween the number of thrips on a sucker and the fol-lowing variables (if not specified, the variables arecontinuous): the height of the sucker; the number ofthrips on the mother plant; the number of thrips inthe neighbourhood; the absence/presence (categoricalvariable) of three weed species [Alocasia cucullata(Aracae), Dieffenbachia seguine (Aracae), Peperomia pel-lucida (Piperacae)]; temperature and rainfall. To testthe influence of the latter variables, we first comparedthe AIC values of models, which differed in theirassessment of the temperature and rainfall means: wecalculated the means for periods beginning at differ-ent times before sampling (from 0 to 28 days) and forperiods of different length (from 1 to 28 days). Wethen selected the temperature and rainfall means forwhich AIC values were the lowest and used meansdetermined in this manner in the complete models totest their significance. We used qGLMs to analyse therelationship between the number of thrips on themother plant and the following variables: the numberof thrips on the sucker; the phenological status (non-flowering or flowering; categorical variable) of themother plant; and the number of thrips in the neigh-bourhood. We used qGLMs to analyse the relation-ship between the number of thrips on the bunch andthe following variables: the number of thrips on themother plant; the presence/absence of a bunch cover(categorical variable); and the number of thrips in theneighbourhood. The qGLMs were selected by follow-ing a backward-stepwise process: we compared thecomplete model to nested submodels using F-tests; wethen removed non-significant fixed effects from thecomplete model and continued the backward-step-wise process until a model was found in which alleffects were significant (Zuur et al. 2009).