
Phylogenetic models of beta and alpha, with their sampling error
Source:R/trait_gradient.R
trait_gradient_phylo.RdFits models of trait evolution to the \(\beta\), \(\alpha\) and mean trait of the
species from trait_gradient(), with the sampling error of each value added to the
phylogenetic covariance, and compares a model without phylogeny, Brownian motion and an
Ornstein-Uhlenbeck process by AIC. It tells whether the position of species along the
gradient (\(\beta\)) and their difference from co-occurring species (\(\alpha\)) are
conserved or labile.
Arguments
- x
Result of
trait_gradient().- tree
Phylogeny (
phylo) with branch lengths; its tips are matched to species by name.- responses
Which of
"beta","alpha"and"mean"to model.- models
Which of
"white","BM"and"OU"to fit.- n_boot
Number of bootstrap samples for the standard errors.
- min_sites
Smallest number of sites of a species with a value (sites where it is alone do not count); species with fewer sites, or not in
tree, are left out.
Value
An object of class mc_trait_gradient_phylo, a list with models (one row per
response and model: response, model, mu, sigma2, a, half_life (\(\ln 2 / a\),
in units of branch length), at_bound (an OU fit with \(a\) at a bound), loglik,
n_par, aic and delta_aic within each response), data (one row per species used: values and standard errors) and tree
(the tree pruned to those species).
Details
The standard error of each value comes from a bootstrap of the sites of each species: the sites (with their trait values and abundances) are resampled with replacement, and \(\bar t_i\), \(\beta_i\) and \(\alpha_i\) recomputed with the site means of the observed data (without the focal species, if so chosen). The site means are held fixed because the question is how well the sites of a species measure its position, not how the gradient itself would change.
For each response \(y\) the model is \(y \sim N(\mu 1, \sigma^2 C + S)\), with
\(S\) the diagonal matrix of squared standard errors and \(C\) the identity
("white"), the shared branch lengths ("BM"), or, for "OU", the covariance of an
Ornstein-Uhlenbeck process with a root at \(\mu\) and strength \(a\),
$$C_{ij} = e^{-a(d_i + d_j - 2 s_{ij})} \frac{1 - e^{-2 a s_{ij}}}{2a},$$
where \(d_i\) is the distance from the root to tip \(i\) and \(s_{ij}\) the length
shared by \(i\) and \(j\). The parameters are estimated by restricted maximum
likelihood, so AIC compares the covariance models with the same mean. Without the error
term, the noise of species seen in few sites passes for evolution near the tips, which an
OU process fits better: in simulations of \(\beta\) under Brownian motion (100
replicates, 30 species, 25 sites), AIC chose OU in 22 without the error term and in 9 with
it.
\(a\) is bounded between \(10^{-3}\) and 50 divided by the height of the tree. At the
lower bound OU is Brownian motion and at the upper bound it is the model without phylogeny;
such fits are marked in at_bound and their half-life is NA. AIC differences among three
models fitted to few species are unstable, and the bootstrap standard error of a species
seen in \(n\) sites is biased low by about \(\sqrt{(n - 1)/n}\).
References
Ackerly, D. D., & Cornwell, W. K. (2007). A trait-based approach to community assembly: partitioning of species trait values into within- and among-community components. Ecology Letters, 10, 135–145. doi:10.1111/j.1461-0248.2006.01006.x
See also
trait_gradient(), trait_ou() to simulate traits under these models.
Other trait variation among individuals:
community_index(),
drift_test(),
neighbour_metrics,
nested_variance(),
null_intervals(),
plot.mc_trait_gradient(),
plot.nested_variance(),
plot.trait_flex(),
plot.trait_rarefaction(),
plot.tstats(),
plot_population_traits(),
population_shifts(),
sample_individuals(),
subsample_individuals(),
trait_densities(),
trait_flex(),
trait_gradient(),
trait_gradient_bayes(),
trait_gradient_test(),
trait_gradient_weights(),
trait_rarefaction(),
tstats(),
tstats_summary()
Examples
set.seed(1)
tree <- treeshapesim::sim_tree_shape(20)
sim <- simulate_metacommunity(landscape_implicit(15), species_traits(tree, lambda = 1),
competition_matrix(tree, "none"), n_steps = 100,
seed_steps = 1:5)
ind <- sample_individuals(sim, n = 5, local_adaptation = 0.5, intraspecific_sd = 1)
tg <- trait_gradient(ind, abundance = "abundance")
trait_gradient_phylo(tg, tree, n_boot = 99)
#> Phylogenetic models of the trait-gradient analysis: 14 species
#> response model mu sigma2 a half_life at_bound loglik n_par aic
#> beta white 64.600 90.9 NA NA NA -49.9 2 104.0
#> beta BM 61.700 201.0 NA NA NA -49.7 2 103.0
#> beta OU 62.900 450.0 1.820 0.38 FALSE -49.2 3 104.0
#> alpha white 0.605 38.7 NA NA NA -44.0 2 92.0
#> alpha BM -1.000 75.6 NA NA NA -41.4 2 86.9
#> alpha OU -1.000 75.7 0.001 NA TRUE -41.4 3 88.9
#> mean white 64.600 236.0 NA NA NA -55.4 2 115.0
#> mean BM 60.000 538.0 NA NA NA -52.9 2 110.0
#> mean OU 60.600 722.0 0.692 1.00 FALSE -52.8 3 112.0
#> delta_aic
#> 0.40
#> 0.00
#> 1.09
#> 5.08
#> 0.00
#> 2.00
#> 4.92
#> 0.00
#> 1.72