Visit, consume and quit: patch quality affects the three stages of foraging
dataset
posted on 2018-06-27, 07:05 authored by Valentina MellaValentina Mella, Malcolm Possell, Sandra Troxell Smith, Clare McArthurClare McArthurWe tested whether the probability of a visit was a function of
treatment (dietary N content as a continuous variable) using logistic regression in SAS (PROC GLIMMIX with a binomial distribution and logit link function, SAS 9.4). Day (fixed effect), site (random effect) and feeding station nested within site (random effect) were also included in the model. We then analysed the effect of treatment (dietary N content as a continuous variable) on visit length (min), each behaviour (% of total time) and GUD (count) separately using the generalized linear mixed model (GLMM) procedure in SAS (PROC GLIMMIX with lognormal distribution and identity link function, SAS 9.4). Day (1-4) was included in the models as a fixed effect, and site and feeding station (nested within site) were random effects.
to determine whether the multivariate VOC data could differentiate the diets along a continuous (dietary nitrogen
content) gradient, similar to analyses of VOCs from other plant/food material. We applied a dispersion weighting followed by square root transformation to the VOC peak area values, then performed CAP analysis on the Bray-Curtis resemblance matrix of the transformed data. To tease apart the contributing VOCs we then applied the CAP
analysis using diet as a class variable. We also isolated the specific volatile signature of the highest quality diet using
the Random Forests (RF). We analysed the data with RF, using a one treatment-versus-the rest approach with the VSURF package (version 1.0.3) in R (version 3.1.2; R Core Team, 2015). Before analysis, TQPA data were transformed using the centred log ratio method using CoDaPack v. 2.01.15.
treatment (dietary N content as a continuous variable) using logistic regression in SAS (PROC GLIMMIX with a binomial distribution and logit link function, SAS 9.4). Day (fixed effect), site (random effect) and feeding station nested within site (random effect) were also included in the model. We then analysed the effect of treatment (dietary N content as a continuous variable) on visit length (min), each behaviour (% of total time) and GUD (count) separately using the generalized linear mixed model (GLMM) procedure in SAS (PROC GLIMMIX with lognormal distribution and identity link function, SAS 9.4). Day (1-4) was included in the models as a fixed effect, and site and feeding station (nested within site) were random effects.
To analyse our VOCs data we looked at the odours of the diets using a canonical analysis of principal coordinates
(CAP) analysis in the PERMANOVA+ add-on of PRIMER v6to determine whether the multivariate VOC data could differentiate the diets along a continuous (dietary nitrogen
content) gradient, similar to analyses of VOCs from other plant/food material. We applied a dispersion weighting followed by square root transformation to the VOC peak area values, then performed CAP analysis on the Bray-Curtis resemblance matrix of the transformed data. To tease apart the contributing VOCs we then applied the CAP
analysis using diet as a class variable. We also isolated the specific volatile signature of the highest quality diet using
the Random Forests (RF). We analysed the data with RF, using a one treatment-versus-the rest approach with the VSURF package (version 1.0.3) in R (version 3.1.2; R Core Team, 2015). Before analysis, TQPA data were transformed using the centred log ratio method using CoDaPack v. 2.01.15.
