Data Tables
Understory Environment
Table 1: An example of the understory environment data table showing some variables measured in the plots.
Understory Vegetation
Table 2: An example of the understory vegetation data table showing percent cover of species.
Examples of the understory environmental and vegetation data collected are presented in Table 1 and 2. For a complete list of environmental variables measured and to see a diagram of the experimental design, refer back to the "Methods" page.
Graphical Data Exploration
Examination of Residuals
Fig. 8: Example of histogram of residuals used to test for normality to meet the assumptions of an ANOVA. Data presented is from the FH layer depth in the control plots of the 1994 stands.
Fig. 9: Example of scatterplot of residuals used to test for homogeneity of variances to meet the assumptions of an ANOVA. Data presented is from the FH layer depth in the control plots of the 1994 stands.
Residuals were created from all environmental data, richness and diversity indexes, and total vegetation cover. These residuals were then tested for normality (Kolmogorov-Smirnov and Shapiro-Wilks) and homogeneity of variance (Bartlett's and Levene's) (see Figures 8 and 9 for examples of plots of the residuals). Soil nutrient data which was not biological relevant (i.e., values 10+ times higher than average which was attributed to contamination) were not included in the data sets during analysis. Data was log transformed when necessary to pass assumptions for an ANOVA. All residuals tested passed at least one test for normality and one test for homogeneity of variance.
Univariate Analysis (ANOVAs)
Univariate analysis (ANOVAs blocked by stand) were used to compare plot locations. Comparisons were made within years and not between years due to different types of ecosites being underplanted in different years. Results from the ANOVAs do not indicate any significant difference (p≤0.05) for the following:
▪ Shannon-Weiner Diversity Index
▪ Whittaker's Diversity Index
▪ Simpson’s Diversity Index
▪ Total plant species richness
▪ Shrub and herb richness
▪ Total percent cover
▪ Total N, NH4N, NO3N, Ca, K, Mg, P
▪ Soil moisture
▪ Litter and FH depth
▪ Decomposition rate
▪ Shannon-Weiner Diversity Index
▪ Whittaker's Diversity Index
▪ Simpson’s Diversity Index
▪ Total plant species richness
▪ Shrub and herb richness
▪ Total percent cover
▪ Total N, NH4N, NO3N, Ca, K, Mg, P
▪ Soil moisture
▪ Litter and FH depth
▪ Decomposition rate
Fig. 10: Boxplot of Shannon's Diversity Index for the various years underplanted/plot locations.
Fig. 11: Boxplot of plot richness for the various years underplanted/plot locations.
Fig. 12: Boxplot of decomposition rate (% loss of cellulose filter paper) for the various years underplanted/plot locations.
Boxplots of raw data presented in Figures 10-12 show examples of species diversity (Shannon's), species richness (total plot) and environmental (decomposition rate) distribution. Similar distributions were observed for the other diversity indexes, richness measures and environmental variables listed above Figure 10 so those boxplots are not presented here.
Multivariate Statistical Analysis
Distance-based Redundancy Analysis (dbRDA)
A distance-based redundancy analysis (dbRDA) can be used to test the significance of individual terms by permutation for a multifactorial experimental design measuring multiple species responses (Legendre and Anderson 1999). It involves first calculating a distance matrix using the distance measure of choice, then running a principal coordinates analysis (PCoA), and finally running a redundancy analysis (RDA) on the eigenvalues obtained from the PCoA. It is useful for ecological data because any distance measure can be used, and because the tests of significance are by permutation, assumptions of normality or homogeneity of variance do not apply.
A dbRDA was used to analyze the environmental and vegetation species data using a Bray-Curtis distance measure (a good measure for species data because of the large number of species absent from some of the plots). This gave an overall view of the environmental and species responses to underplanting.
A distance-based redundancy analysis (dbRDA) can be used to test the significance of individual terms by permutation for a multifactorial experimental design measuring multiple species responses (Legendre and Anderson 1999). It involves first calculating a distance matrix using the distance measure of choice, then running a principal coordinates analysis (PCoA), and finally running a redundancy analysis (RDA) on the eigenvalues obtained from the PCoA. It is useful for ecological data because any distance measure can be used, and because the tests of significance are by permutation, assumptions of normality or homogeneity of variance do not apply.
A dbRDA was used to analyze the environmental and vegetation species data using a Bray-Curtis distance measure (a good measure for species data because of the large number of species absent from some of the plots). This gave an overall view of the environmental and species responses to underplanting.
Non-linear Multidimensional Scaling (NMDS) Ordinations
Non-linear multidimensional scaling (NMDS) ordinates observations based upon a matrix of similarity created using a distance measure of choice; observations plotted closer together are more similar. NMDS uses ranked distances to perform an iterative search for positions which best minimize stress. This analysis avoids assumptions of linear relationship among variables and normality, making it useful for ecological data. NMDS also performs well even when beta diversity is high (McCune and Grace 2002).
A NMDS (using a Bray-Curtis distance measure) was used to analyze the vegetation data by cover type to see if there were differences in vegetation cover type among plot locations due to underplanting (since vegetation cover type has been shown to differ among braodleaf, mixedwood and conifer stands by Macdonald and Fenniak (2007)).
Non-linear multidimensional scaling (NMDS) ordinates observations based upon a matrix of similarity created using a distance measure of choice; observations plotted closer together are more similar. NMDS uses ranked distances to perform an iterative search for positions which best minimize stress. This analysis avoids assumptions of linear relationship among variables and normality, making it useful for ecological data. NMDS also performs well even when beta diversity is high (McCune and Grace 2002).
A NMDS (using a Bray-Curtis distance measure) was used to analyze the vegetation data by cover type to see if there were differences in vegetation cover type among plot locations due to underplanting (since vegetation cover type has been shown to differ among braodleaf, mixedwood and conifer stands by Macdonald and Fenniak (2007)).
Principal Component Analysis (PCA)
Principal component analysis (PCA) uses a Euclidean distance measure to reduce a data set containing many variables into a few composite variables, which represent most of the original data and show the strongest covariation among variables in the first few components (McCune and Grace 2002). PCA assumes a linear relationship among variables so the assumptions of normality must be met.
A PCA was used to analyze the environmental data set to see if there were differences in the understory environment at difference plot locations due to underplanting. This analysis was done because if no differences were observed in the understory environment, there would be no reason to suspect changes to the understory vegetation; this would indicate white spruce was not an ecosystem engineer in these underplanted stands.
Principal component analysis (PCA) uses a Euclidean distance measure to reduce a data set containing many variables into a few composite variables, which represent most of the original data and show the strongest covariation among variables in the first few components (McCune and Grace 2002). PCA assumes a linear relationship among variables so the assumptions of normality must be met.
A PCA was used to analyze the environmental data set to see if there were differences in the understory environment at difference plot locations due to underplanting. This analysis was done because if no differences were observed in the understory environment, there would be no reason to suspect changes to the understory vegetation; this would indicate white spruce was not an ecosystem engineer in these underplanted stands.
Permutational Multivariate Analysis of Variance (perMANOVA)
Permutational multivariate analysis of variance (perMANOVA) is a non-parametic procedure which allows testing of the effects of factors and their interactions by permutation in multivariate datasets which do not pass the assumptions of normality required to run a MANOVA (Anderson 2001).
A perMANOVA was used to test the significance of the year underplanted and of the plot locations for the environmental and species data. A perMANOVA was used because the species data did not meet the assumptions for a MANOVA.
Permutational multivariate analysis of variance (perMANOVA) is a non-parametic procedure which allows testing of the effects of factors and their interactions by permutation in multivariate datasets which do not pass the assumptions of normality required to run a MANOVA (Anderson 2001).
A perMANOVA was used to test the significance of the year underplanted and of the plot locations for the environmental and species data. A perMANOVA was used because the species data did not meet the assumptions for a MANOVA.
Disclaimer: This website was created as part of a class exercise for RenR 690 by Erica Graham.