modEvA (version 3.33)

varPart: Variation partitioning

Description

This function performs variation partitioning (Borcard et al. 1992) among two factors (e.g. Ribas et al. 2006) or three factors (e.g. Real et al. 2003) for either linear regression models (LM) or generalized linear models (GLM).

Usage

varPart(A, B, C = NA, AB, AC = NA, BC = NA, ABC = NA, model.type = NULL,
A.name = "Factor A", B.name = "Factor B", C.name = "Factor C",
model = NULL, groups = NULL, pred.type = "Y", cor.method = "pearson",
return.models = FALSE, plot = TRUE, plot.digits = 3, cex.names = 1.5,
cex.values = 1.2, main = "", cex.main = 2, plot.unexpl = TRUE, colr = FALSE)

Value

This function returns a data frame indicating the proportion of variance accounted for by each of the factors or groups, and (if 'plot = TRUE') a Venn diagram of the contributions of each factor or overlap. If 'return.models=TRUE', the output includes also the model obtained for each group of variables.

Arguments

A

numeric value of the R-squared of the regression of the response variable on the variables related to factor 'A'. NOTE: INSTEAD of this and the next 10 arguments, you can use arguments 'model' and 'groups' below.

B

numeric value of the R-squared of the regression of the response variable on the variables related to factor 'B'

C

(optionally, if there are 3 factors) numeric value of the R-squared of the regression of the response on the variables related to factor 'C'

AB

numeric value of the R-squared of the regression of the response on the variables of factors 'A' and 'B' simultaneously

AC

(if there are 3 factors) numeric value of the R-squared of the regression of the response on the variables of factors 'A' and 'C' simultaneously

BC

(if there are 3 factors) numeric value of the R-squared of the regression of the response on the variables of factors 'B' and 'C' simultaneously

ABC

(if there are 3 factors) numeric value of the R-squared of the regression of the response on the variables of factors 'A', 'B' and 'C' simultaneously

model.type

deprecated argument, kept here for back-compatibility

A.name

character string indicating the name of factor 'A'

B.name

character string indicating the name of factor 'B'

C.name

character string indicating the name of factor 'C' (if there are 3 factors)

model

a model object of class 'glm' (for linear models, instead of 'lm' you can use use 'glm' with family=gaussian). If this argument is provided, all previous arguments are ignored, as they are computed instead from 'model' and 'groups'.

groups

data frame with 2 columns, the 1st one containing the names of the variables, and the 2nd one containing the names of the factors in which they should be grouped (e.g. climatic, human, topographic) for the variation partitioning. This argument is required (and only used) if 'model' is provided.

pred.type

character value specifying the type of predictions among which to calculate the R-squared values, or squared correlations. Can be "Y" (the default) for the 'link' function (in the scale of the predictor variables), "P" for using the 'response' (e.g. in the scale of probability for models of family binomial), or "F" for using favourability (i.e., probability after removing the effect of modelled prevalence, as in the 'Fav' function of package fuzzySim); see Details. This argument is only used if 'model' is provided.

cor.method

character value to pass to the 'method' argument of cor specifying the correlation coefficient to use. The default is "pearson". This argument is only used if 'model' is provided.

return.models

logical value indicating whether to include in the output the model obtained for each group of variables. The default is FALSE. This argument is only used if 'model' is provided.

plot

logical, whether to plot the variation partitioning diagram. The default is TRUE.

plot.digits

integer value of the number of digits to which to round the values in the plot. The default is 3.

cex.names

numeric value indicating character expansion factor to define the size of the names of the factors displayed in the plot.

cex.values

numeric value indicating character expansion factor to define the size of the values displayed in the plot.

main

optional character string indicating the main title for the plot. The default is empty.

cex.main

numeric value indicating character expansion factor to define the font size of the plot title (if provided).

plot.unexpl

logical value indicating whether the amount of unexplained variation should be included in the plot. The default is TRUE.

colr

logical value indicating whether or not to colour the circles in the plot. The default is FALSE for back-compatibility.

Author

A. Marcia Barbosa

Details

If you have linear models (i.e. GLMs of family Gaussian), input data for 'varPart' are the coefficients of determination (R-squared values) of the linear regressions of the response variable on all the variables in the model, on the variables related to each particular factor, and (when there are 3 factors) on the variables related to each pair of factors. The outputs are the amounts of variance explained exclusively by each factor, the amounts explained exclusively by the overlapping effects of each pair of factors, and the amount explained by the overlap of the 3 factors if this is the case (e.g. Real et al. 2003). The amount of variation not explained by the complete model is also provided.

If you have generalized linear models (GLMs) such as logistic regression (see glm), you have no true R-squared values; inputs can then be the squared coefficients of correlation between the model predictions given by each factor (or pair of factors) and the predictions of the complete model. Predictions can be probability (e.g. Munoz & Real 2006), favourability (Baez et al. 2012, Estrada et al. 2016), or the 'logit' linear predictor (Real et al. 2013); the correltion coefficient can be e.g. Pearson's (Munoz & Real 2006) or Spearman's (Baez et al. 2012). An adjusted R-squared can also be used (De Araujo et al. 2014). In GLMs, the "total variation" (AB or ABC, depending on whether you have two or three factors) is 1 (correlation of the predictions of the complete model with themselves), and output values are not the total amounts of variance (of the response variable) explained by variable groups and their overlaps, but rather their proportional contribution to the total variation explained by the model.

References

Baez J.C., Estrada A., Torreblanca D. & Real R (2012) Predicting the distribution of cryptic species: the case of the spur-thighed tortoise in Andalusia (southern Iberian Peninsula). Biodiversity and Conservation 21: 65-78

Borcard D., Legendre P., Drapeau P. (1992) Partialling out the spatial component of ecological variation. Ecology 73: 1045-1055

De Araujo C.B., Marcondes-Machado L.O. & Costa G.C. (2014) The importance of biotic interactions in species distribution models: a test of the Eltonian noise hypothesis using parrots. Journal of Biogeography 41: 513-523

Estrada A., Delgado M.P., Arroyo B., Traba J., Morales M.B. (2016) Forecasting Large-Scale Habitat Suitability of European Bustards under Climate Change: The Role of Environmental and Geographic Variables. PLoS ONE 11(3): e0149810

Munoz A.-R. & Real R. (2006) Assessing the potential range expansion of the exotic monk parakeet in Spain. Diversity and Distributions 12: 656-665

Real R., Barbosa A.M., Porras D., Kin M.S., Marquez A.L., Guerrero J.C., Palomo L.J., Justo E.R. & Vargas J.M. (2003) Relative importance of environment, human activity and spatial situation in determining the distribution of terrestrial mammal diversity in Argentina. Journal of Biogeography 30: 939-947

Real R., Romero D., Olivero J., Estrada A. & Marquez A.L. (2013) Estimating how inflated or obscured effects of climate affect forecasted species distribution. PLoS ONE 8: e53646

Ribas A., Barbosa A.M., Casanova J.C., Real R., Feliu C. & Vargas J.M. (2006) Geographical patterns of the species richness of helminth parasites of moles (Talpa spp.) in Spain: separating the effect of sampling effort from those of other conditioning factors. Vie et Milieu 56: 1-8

Examples

# if you have a linear model (LM), use (non-adjusted) R-squared values
# for each factor and for their combinations as inputs:

# with 2 factors:

varPart(A = 0.456, B = 0.315, AB = 0.852, A.name = "Spatial",
B.name = "Environmental", main = "Small whale")

varPart(A = 0.456, B = 0.315, AB = 0.852, A.name = "Spatial",
B.name = "Environmental", main = "Small whale", colr = TRUE)


# with 3 factors:

varPart(A = 0.456, B = 0.315, C = 0.281, AB = 0.051, BC = 0.444,
AC = 0.569, ABC = 0.624, A.name = "Spatial", B.name = "Human",
C.name = "Environmental", main = "Small whale")

varPart(A = 0.456, B = 0.315, C = 0.281, AB = 0.051, BC = 0.444,
AC = 0.569, ABC = 0.624, A.name = "Spatial", B.name = "Human",
C.name = "Environmental", main = "Small whale", colr = TRUE)


# if you have a generalized linear model (GLM),
# you can use squared Pearson correlation coefficients of the
# predictions of each factor with those of the complete model:

varPart(A = (-0.005)^2, B = 0.698^2, C = 0.922^2, AB = 0.696^2,
BC = 0.994^2, AC = 0.953^2, ABC = 1, A.name = "Topographic",
B.name = "Climatic", C.name = "Geographic", main = "Big bird")

# but "Unexplained variation" can be deceiving in these cases
# (see Details); try also adding 'plot.unexpl = FALSE'


# if you have a model object and a table classifying the variables into groups:

data(rotif.mods)
mod <- rotif.mods$models[[2]]

head(mod$model)
vars <- colnames(mod$model)[-1]
vars
var_groups <- data.frame(vars = vars, groups = c("Spatial", "Spatial",
"Climate", "Climate", "Climate", "Human"))
var_groups

varPart(model = mod, groups = var_groups)
varPart(model = mod, groups = var_groups, pred.type = "P", colr = TRUE)