The sequence of a small part of the insertion could not be determined, probably because of the presence of high-dimensional structure. Statistical analyses for the field census and insect-removal experiment were conducted using R 2.7.1 . The effect of trichome phenotype on fruit production in the field census was LY2157299 msds examined by fitting LY2835219 generalized linear models with negative binomial error . The trichome phenotype, rosette diameter , and census plot were included as fixed factors. A GLM with negative binomial error was used because many plants produced no fruits, and the frequency distribution of fruit production was strongly overdispersed from the Poisson distribution . The number of P. brassicae larvae on plants during the flowering season was examined using a generalized linear mixed effect model with Poisson error in which the trichome phenotype and plot were included as fixed factors and repeated measurements on each plant as a random factor. AICs were compared for models with and without the trichome term. Insects other than P. brassicae larvae showed low infestation rates and were not included in the analyses. In the insect-removal experiment, the effects of the trichome phenotype and insecticide treatment on fruit production were examined by GLMM with negative binomial error . The trichome phenotype, treatment, rosette diameter, and transplanting plot were included as fixed factors and the maternal plant as a random factor. The number of P. brassicae larvae on plants during the flowering season in the transplanting experiment was examined by GLMM with Poisson error in which the trichome phenotype, insecticide treatment, and plot were included as fixed factors and repeated measurements on each individual plant as a random factor. AICs were calculated for the full model and simplified models in which the trichome and treatment terms were sequentially subtracted. For plants faced only with autumn herbivory, GLM with negative binomial error was fitted to examine the effects of the treatment and trichome phenotype on flower production. Rosette diameter and transplanting plot were also included as covariates. To investigate how the pattern of GL1 polymorphism differs from that of the two adjacent genes, Tajima��s D and Fu & Li��s D and F were calculated for the three genes using DnaSP 4.20 software .