DIGITALNA ARHIVA ŠUMARSKOG LISTA
prilagođeno pretraživanje po punom tekstu
|ŠUMARSKI LIST 5-6/2022 str. 52 <-- 52 --> PDF|
Additionally, patch metrics, environmental estimates together with human pressure variables were tested with binomial multivariate logistic regression models to describe the presence of P. serotina and I. parviflora related to patch characteristics, using the glm function with binomial family distribution. In the full models, all variables shown in Table 1 were included. Continuous variables were z-standardized before the models were implemented to allow comparison of variable effects. Regression coefficients were transformed using the exponential function to yield the odds ratio associated with a one-unit increase in the case of continuous variables or comparison with the baseline level in the case of factors.
The best model (final model) was selected according to the lowest AIC value with the dredge function in the “MuMIn” package (Barton, 2009). The absence of multicollinearity among the explanatory variables included in the final model was determined by calculating the variance inflation factor (VIF < 2; Fox and Monette, 1992) in the “car” package (Fox and Weisber, 2019). Goodness of model fit was evaluated with Chi-square test (Χ2) statistics at a significance level P < 0.05. Χ2 was calculated as the deviance of the null model subtracted from the deviance of the model fit (Δ deviance) and degrees of freedom of the null model subtracted from degrees of freedom of model fit. Statistical analyses were conducted in the statistical software environment RStudio for R (Version 1.3.1093 for Windows; R Core Team, 2019).
Prunus serotina and Impatiens parviflora were detected in 25% and 48% of studied forest patches, respectively. Comparison between forest patches with P. serotina present or absent revealed significant differences in P/A ratios and estimated nutrients and light conditions (Fig. 1). On the other hand, we did not find differences between forest patches where I. parviflora was present and those where it was absent (Fig. 1). Nevertheless, with further analyses using logistic regression we detected some interesting trends discussed below.
Variables that significantly affected the probability of each species being present in the forest patch were species specific. Comparing alternative multivariate models predicting the presence of each species (Appendix 1 and Appendix 2)