Construct a multimodel species-area relationship curve using information criterion weights and up to twenty SAR models.
sar_multi(data, obj = c("power",
"powerR","epm1","epm2","p1","p2","expo","koba",
"mmf","monod","negexpo","chapman","weibull3","asymp",
"ratio","gompertz","weibull4","betap","heleg", "linear"), crit = "Info",
normaTest = "lillie", homoTest = "cor.fitted", neg_check = TRUE,
alpha_normtest = 0.05, alpha_homotest = 0.05, confInt = FALSE, ciN = 100,
verb = TRUE)A dataset in the form of a dataframe with two columns: the first with island/site areas, and the second with the species richness of each island/site.
Either a vector of model names or a fit_collection object.
The criterion used to compare models and compute the model
weights. The default crit = "Info" switches to AIC or AICc depending
on the number of data points in the dataset. For BIC, use crit =
"Bayes".
The test used to test the normality of the residuals of each model. Can be any of "lillie" (Lilliefors Kolmogorov-Smirnov test; the default), "shapiro" (Shapiro-Wilk test of normality), "kolmo" (Kolmogorov-Smirnov test), or "none" (no residuals normality test is undertaken).
The test used to check for homogeneity of the residuals of each model. Can be any of "cor.fitted" (a correlation of the residuals with the model fitted values; the default), "cor.area" (a correlation of the residuals with the area values), or "none" (no residuals homogeneity test is undertaken).
Whether or not a check should be undertaken to flag any models that predict negative richness values.
The alpha value used in the residual normality test (default = 0.05, i.e. any test with a P value < 0.05 is flagged as failing the test).
The alpha value used in the residual homogeneity test (default = 0.05, i.e. any test with a P value < 0.05 is flagged as failing the test).
A logical argument specifying whether confidence intervals should be calculated for the multimodel curve using bootstrapping.
The number of bootstrap samples to be drawn to calculate the
confidence intervals (if confInt == TRUE).
verbose (default: verb == TRUE).
A list of class "multi" and class "sars" with two elements. The first element ('mmi') contains the fitted values of the multimodel sar curve. The second element ('details') is a list with the following components:
mod_names Names of the models that were successfully fitted and passed any model check
fits A fit_collection object containing the successful model fits
ic The information criterion selected
norm_test The residual normality test selected
homo_test The residual homogeneity test selected
alpha_norm_test The alpha value used in the residual normality test
alpha_homo_test The alpha value used in the residual homogeneity test
ics The information criterion values (e.g. AIC values) of the model fits
delta_ics The delta information criterion values
weights_ics The information criterion weights of each model fit
n_points Number of data points
n_mods The number of successfully fitted models
no_fit Names of the models which could not be fitted or did not pass model checks
The summary.sars function returns a more useful summary of
the model fit results, and the plot.multi plots the
multimodel curve.
The multimodel SAR curve is constructed using information criterion
weights (see Burnham & Anderson, 2002; Guilhaumon et al. 2010). If
obj is a vector of n model names the function fits the n models to
the dataset provided. A dataset must have four or more datapoints to fit
the multimodel curve. If any models cannot be fitted they are removed from
the multimodel SAR. If obj is a fit_collection object, any model
fits in the collection which are NA are removed. In addition, if any other
model checks have been selected (i.e. residual normality and heterogeneity
tests, and checks for negative predicted richness values), these are
undertaken and any model that fails the selected test(s) is removed from
the multimodel SAR. The order of the additional checks inside the function
is: normality of residuals, homogeneity of residuals, and a check for
negative fitted values. Once a model fails one test it is removed and thus
is not available for further tests. Thus, a model may fail multiple tests
but the returned warning will only provide information on a single test.
The resultant models are then used to construct the multimodel SAR curve.
For each model in turn, the model fitted values are multiplied by the
information criterion weight of that model, and the resultant values are
summed across all models (Burnham & Anderson, 2002). Confidence intervals
can be calculated (using confInt) around the multimodel averaged
curve using the bootstrap procedure outlined in Guilhaumon et al (2010).The
procedure transforms the residuals from the individual model fits and
occasionally NAs / Inf values can be produced - in these cases, the model
is removed from the confidence interval calculation (but not the multimodel
curve itself). When several SAR models are used and the number of bootstraps
(ciN) is large, generating the confidence intervals can take a long
time.
Burnham, K. P., & Anderson, D. R. (2002). Model selection and multi-model inference: a practical information-theoretic approach (2nd ed.). New-York: Springer.
Guilhaumon, F., Mouillot, D., & Gimenez, O. (2010). mmSAR: an R-package for multimodel species-area relationship inference. Ecography, 33, 420-424.
# NOT RUN {
data(galap)
#attempt to construct a multimodel SAR curve using all twenty sar models
fit <- sar_multi(galap)
summary(fit)
plot(fit)
# construct a multimodel SAR curve using a fit_collection object
s1 <- sar_power(galap)
s2 <- sar_expo(galap)
s3 <- sar_koba(galap)
ff <- fit_collection(s1, s2, s3)
fit2 <- sar_multi(galap, obj = ff)
summary(fit2)
# construct a multimodel SAR curve without conducting any model checks
fit3 <- sar_multi(galap, normaTest = "none", homoTest = "none", neg_check = FALSE)
# }
Run the code above in your browser using DataLab