To analyze the effect of landscape composition and configuration, and effective population size on genetic variability and differentiation we first analyzed the scale of effect for each response variable (Jackson and Fahrig, 2012), using multiscale test of independence for multivariate vectors implemented in the multifit function (Huais, 2018), in R version 3.6.1 (R Core Team, 2019). We used linear models and R-squared (R2) as an index of the strength of the relationship (see the respective scale of effect to each variable in Supplementary Table 12).

Then, for the variables selected using the scale of effect (Supplementary Table 12), we analyzed the collinearity among the explanatory variables by estimating the variance inflation factor (VIF), which measures the inflation of the variance of a regression coefficient caused by multicollinearity in the model (Dormann et al., 2013). Analyses were performed using the jtools package (Long, 2019) in R version 3.6.1, with a stepwise approach to eliminate models with VIFs > 5.0 (Zuur et al., 2010). The analyses were performed for each response variable (see Supplementary Table 13).

To identify the effects of landscape on adaptive and neutral variation (see predictions in Figure 1), we then performed linear models using the explanatory variables selected by scale of effect and multicollinearity analysis (VIF ≤ 5.0, see Table 1). The response variables are continuous and we assumed Gaussian distributions on model fitting. We also built a null model by randomly sampling data keeping β equal to zero (constant variables) for all explanatory variables (absence of specific landscape processes). At node level we select the best predictive model based on Akaike Information Criteria (AIC). We estimated AIC corrected for small sample sizes (AICc), i.e., the difference of each model and the best model (ΔAICci), and Akaike’s Weight of Evidence (wAICc), i.e., the relative contribution of each model to explain the observed pattern, given a set of competing models (Burnham and Anderson, 2002). Models with ΔAICc < 2 were considered as equally plausible to explain the patterns (Zuur et al., 2009). For link level, we used models significance to select the best predictive model, because of the small sample size (five landscapes). All analyses were carried out using the packages mgvc (Wood, 2020), bblme (Bolker and R Development Core Team, 2017), visreg (Breheny and Burchett, 2017) and ggplot2 (Wickham, 2016), available in R version 3.6.1 (R Core Team, 2019).

Models performed at node and link levels for both neutral and adaptive quantitative traits measured in seeds and seedlings of Caryocar brasiliense.

Note: The content above has been extracted from a research article, so it may not display correctly.

Please log in to submit your questions online.
Your question will be posted on the Bio-101 website. We will send your questions to the authors of this protocol and Bio-protocol community members who are experienced with this method. you will be informed using the email address associated with your Bio-protocol account.

We use cookies on this site to enhance your user experience. By using our website, you are agreeing to allow the storage of cookies on your computer.