For graphical displays, we retained the finer categories recorded. Only the features with significant effects between pairs of herbivores www.selleckchem.com/products/Fludarabine(Fludara).html were considered in building the full models. Because our measures of woody vegetation structure (tree cover, shrub cover and tree height) were all positively correlated (Spearman

rank correlation r > 0.4), we retained only tree cover in the full model. The relative support for models incorporating different combinations of habitat features was assessed using the Akaike Information Criterion (AIC; Burnham & Anderson, 2002). Following Agresti (1990), for categorical data, AIC = G2 – 2d.f., where G2 is the deviance of each model or log likelihood ratio chi-square, and d.f. refers to degrees of freedom. To assess selection of grass species, we considered each feeding site (encompassing nine quadrats) to be an independent sample of food choice. Following Owen-Smith & Cooper (1987),

the availability of each grass species was calculated as the proportion of feeding sites where each species was present within one or more quadrats. The site-based acceptance of each grass species present at ≥10 LBH589 cost feeding sites for each herbivore species was obtained by dividing the number of feeding sites where each grass species was grazed in at least one quadrat by the total number of feeding sites where the grass species was present. Grass greenness was categorized as follows: <5%, 6–10%, 11–20% and >20%. For grass height, only two categories were used, because too few grass tufts were shorter than 20 cm or taller than 80 cm in the feeding sites. We also calculated acceptance ratings for grass tufts aggregated into greenness categories independently of the grass species represented for each herbivore species. Log-linear analysis was used to compare structural and phenological

features of the grass tufts grazed by sable with those grazed by buffalo or zebra. The change in model fit when the factor herbivore 上海皓元医药股份有限公司 was removed from the model was evaluated using the likelihood ratio test (G2) for P < 0.05. We examined z-scores to identify the cells of the contingency table that contributed most to the lack of fit of the reduced model, thereby identifying the main distinctions in grass species availability or acceptability between the herbivore species. To estimate the dietary contribution by each grass species for each grazer, we assumed morning or afternoon foraging periods to be independent sampling units. The proportional dietary contribution was estimated by dividing the number of bites taken on tufts of each grass species by the total number of bites recorded at the feeding sites. The resulting proportions were averaged across foraging periods to obtain the monthly or seasonal diet contribution by each grass species.