Compute STRAPP to test for a relationship between diversification rates and trait data

Carries out the appropriate statistical method to test for a relationship between diversification rates and trait data for a given point in the past (i.e. the focal_time). Tests are based on block-permutations: rates data are randomized across tips following blocks defined by the diversification regimes identified on each tip (typically from a BAMM).

Such tests are called STructured RAte Permutations on Phylogenies (STRAPP) as described in Rabosky, D. L., & Huang, H. (2016). A robust semi-parametric test for detecting trait-dependent diversification. Systematic biology, 65(2), 181-193. https://doi.org/10.1093/sysbio/syv066.

The function is an extension of the original BAMMtools::traitDependentBAMM() function used to carry out STRAPP test on extant time-calibrated phylogenies.

Tests can be carried out on speciation, extinction and net diversification rates.

deepSTRAPP::compute_STRAPP_test_for_focal_time() can handle three types of statistical tests depending on the type of trait data provided:

Continuous trait data

Tests for correlations between trait and rates carried out with deepSTRAPP::compute_STRAPP_test_for_continuous_data(). The associated test is the Spearman's rank correlation test (See stats::cor.test).

Binary trait data

For categorical and biogeographic trait data that have only two states (ex: 'Nearctic' vs. 'Neotropics'). Tests for differences in rates between states are carried out with deepSTRAPP::compute_STRAPP_test_for_binary_data(). The associated test is the Mann-Whitney-Wilcoxon rank-sum test (See stats::wilcox.test).

Multinominal trait data

For categorical and biogeographic trait data with more than two states (ex: 'No leg' vs. 'Two legs' vs. 'Four legs'). Tests for differences in rates between states are carried out with deepSTRAPP::compute_STRAPP_test_for_multinominal_data(). The associated test for all states is the Kruskal-Wallis H test (See stats::kruskal.test). If posthoc_pairwise_tests = TRUE, post hoc pairwise tests between pairs of states will be carried out too. The associated test for post hoc pairwise tests is the Dunn's post hoc pairwise rank-sum test (See dunn.test::dunn.test).

Usage

compute_STRAPP_test_for_focal_time(
  BAMM_object,
  trait_data_list,
  rate_type = "net_diversification",
  seed = NULL,
  nb_permutations = NULL,
  replace_samples = FALSE,
  alpha = 0.05,
  two_tailed = TRUE,
  one_tailed_hypothesis = NULL,
  posthoc_pairwise_tests = FALSE,
  p.adjust_method = "none",
  return_perm_data = FALSE,
  nthreads = 1,
  print_hypothesis = TRUE
)

Arguments

BAMM_object

Object of class "bammdata", typically generated with update_rates_and_regimes_for_focal_time(), that contains a phylogenetic tree and associated diversification rates across selected posterior samples updated to a specific time in the past (i.e. the focal_time).

trait_data_list

List obtained from extract_most_likely_trait_values_for_focal_time() that contains at least a $trait_data element, a $focal_time element, and a $trait_data_type. $trait_data is a named vector with the trait data found on the phylogeny at focal_time. $focal_time informs on the time in the past at which the trait and rates data will be tested. $trait_data_type informs on the type of trait data: continuous, categorical, or biogeographic.

rate_type

A character string specifying the type of diversification rates to use. Must be one of 'speciation', 'extinction' or 'net_diversification' (default).

seed

Integer. Set the seed to ensure reproducibility. Default is NULL (a random seed is used).

nb_permutations

Integer. To select the number of random permutations to perform during the tests. If NULL (default), all posterior samples will be used once.

replace_samples

Logical. To specify whether to allow 'replacement' (i.e., multiple use) of a posterior sample when drawing samples used to carry out the test. Default is FALSE.

alpha

Numerical. Significance level to use to compute the estimate corresponding to the values of the test statistic used to assess significance of the test. This does NOT affect p-values. Default is 0.05.

two_tailed

Logical. To define the type of tests. If TRUE (default), tests for correlations/differences in rates will be carried out with a null hypothesis that rates are not correlated with trait values (continuous data) or equals between trait states (categorical and biogeographic data). If FALSE, one-tailed tests are carried out.

For continuous data, it involves defining a one_tailed_hypothesis testing for either a "positive" or "negative" correlation under the alternative hypothesis.
For binary data (two states), it involves defining a one_tailed_hypothesis indicating which states have higher rates under the alternative hypothesis.
For multinominal data (more than two states), it defines the type of post hoc pairwise tests to carry out between pairs of states. If posthoc_pairwise_tests = TRUE, all two-tailed (if two_tailed = TRUE) or one-tailed (if two_tailed = FALSE) tests are automatically carried out.

one_tailed_hypothesis

A character string specifying the alternative hypothesis in the one-tailed test. For continuous data, it is either "negative" or "positive" correlation. For binary data, it lists the trait states with states ordered in increasing rates under the alternative hypothesis, separated by a greater-than such as c('A > B').

posthoc_pairwise_tests

Logical. Only for multinominal data (with more than two states). If TRUE, all possible post hoc pairwise (Dunn) tests will be computed across all pairs of states. This is a way to detect which pairs of states have significant differences in rates if the overall test (Kruskal-Wallis) is significant. Default is FALSE.

p.adjust_method

A character string. Only for multinominal data (with more than two states). It specifies the type of correction to apply to the p-values in the post hoc pairwise tests to account for multiple comparisons. See stats::p.adjust() for the available methods. Default is none.

return_perm_data

Logical. Whether to return the stats data computed from the posterior samples for observed and permuted data in the output. This is needed to plot the histogram of the null distribution used to assess significance of the test with plot_histogram_STRAPP_test_for_focal_time(). Default is FALSE.

nthreads

Integer. Number of threads to use for paralleled computing of the tests across the permutations. The R package parallel must be loaded for nthreads > 1. Default is 1.

print_hypothesis

Logical. Whether to print information on what test is carried out, detailing the null and alternative hypotheses, and what significant level is used to rejected or not the null hypothesis. Default is TRUE.

Value

The function returns a list with at least eight elements.

Summary elements for the main test:

$estimate Named numerical. Value of the test statistic used to assess significance of the test according to the significance level provided (alpha). The test is significant if $estimate is higher than zero.
$stats_median Numerical. Median value of the distribution of test statistics across all selected posterior samples.
$p-value Numerical. P-value of the test. The test is considered significant if $p-value is lower than alpha.
$method Character string. The statistical method used to carry out the test.
$rate_type Character string. The type of diversification rates tested. One of 'speciation', 'extinction' or 'net_diversification'.
$trait_data_type Character string. The type of trait data as found in 'trait_data_list$trait_data_type'. One of 'continuous', 'categorical', or 'biogeographic'.
$trait_data_type_for_stats Character string. The type of trait data used to select statistical method. One of 'continuous', 'binary', or 'multinominal'.
$focal_time The time in the past at which the trait and rates data were tested.

If using continuous or binary data:

$two-tailed Logical. Record the type of test used: two-tailed if TRUE, one-tailed if FALSE. If one_tailed_hypothesis is provided (only for continuous and binary trait data):
$one_tailed_hypothesis Character string. Record of the alternative hypothesis used for the one-tailed tests.

If posthoc_pairwise_tests = TRUE (only for multinomial trait data):

$posthoc_pairwise_tests List of at least 3 sub-elements:
- $summary_df Data.frame of five variables providing the summary results of post hoc pairwise tests
- $method Character string. The statistical method used to carry out the test. Here, "Dunn".
- $two-tailed Logical. Record the type of post hoc pairwise tests used: two-tailed if TRUE, one-tailed if FALSE.

If return_perm_data = TRUE, the stats data computed from the posterior samples for observed and permuted data are provided. This is needed to plot the histogram of the null distribution used to assess significance of the test with plot_histogram_STRAPP_test_for_focal_time().

$perm_data_df A data.frame with four variables summarizing the data generated during the STRAPP test:
- $posterior_samples_random_ID Integer. ID of the posterior samples randomly drawn and used for the STRAPP test.
- $*_obs Numerical. Test stats computed from the observed data in the posterior samples. Name depends on the test used.
- $*_perm Numerical. Test stats computed from the permuted data in the posterior samples. Name depends on the test used.
- $delta_* OR $abs_delta_* Numerical. Test stats computed for the STRAPP test comparing observed stats and permuted stats. Name depends on the test used and the type of tests (two-tailed compare absolute values; one-tailed compare raw values). Combined with posthoc_pairwise_tests = TRUE, the stats data are also provided for the post hoc pairwise tests:
$posthoc_pairwise_tests$perm_data_array A 3D array containing stats data for all post hoc pairwise tests in a similar format that $perm_data_df.

If no STRAPP test was performed in the case of categorical/biogeographic data with a single state/range at focal_time, only the $trait_data_type, $trait_data_type_for_stats = "none", and $focal_time are returned.

Details

These set of functions carries out the STructured RAte Permutations on Phylogenies (STRAPP) test as defined in Rabosky, D. L., & Huang, H. (2016). A robust semi-parametric test for detecting trait-dependent diversification. Systematic biology, 65(2), 181-193.

It is an extension of the original BAMMtools::traitDependentBAMM() function used to carry out STRAPP test on extant time-calibrated phylogenies, but allowing here to test for differences/correlations at any point in the past (i.e. the focal_time).

It takes an object of class "bammdata" (BAMM_object) that was updated such as its diversification rates ($tipLambda and $tipMu) and regimes ($tipStates) are reflecting values observed at at a specific time in the past (i.e. the $focal_time). Similarly, it takes a list (trait_data_list) that provides $trait_data as observed on branches at the same focal_time than the diversification rates and regimes.

A STRAPP test is carried out by drawing a random set of posterior samples from the BAMM_object, then randomly permuting rates across blocks of tips defined by the macroevolutionary regimes. Test statistics are then computed across the initial observed data and the permuted data for each sample. In a two-tailed test, the p-value is the proportion of posterior samples in which the test stats is as extreme in the permuted than in the observed data. In a one-tailed test, the p-value is the proportion of posterior samples in which the test stats is higher in the permuted than in the observed data.

———- Major changes compared to BAMMtools::traitDependentBAMM() ———-

Allow to choose if random sampling of posterior configurations must be done with replacement or not with replace_samples.
Add post hoc pairwise tests (Dunn test) for multinominal data. Use posthoc_pairwise_tests = TRUE.
Provide outputs tailored for histogram plots plot_histogram_STRAPP_test_for_focal_time() and p-value time-series plots plot_STRAPP_pvalues_over_time().
Add prints detailing what test is carried out, what are the null and alternative hypotheses, and what significant level is used to rejected or not the null hypothesis. (Enabled with print_hypothesis = TRUE).
Split the function in multiple sub-functions according to the type of data ($trait_data_type).
Prevent using Pearson's correlation tests and applying log-transformation for continuous data. The rationale is that there is no reason to assume that tip rates are distributed normally or log-normally. Thus, a Spearman's rank correlation test is favored.

References

For STRAPP: Rabosky, D. L., & Huang, H. (2016). A robust semi-parametric test for detecting trait-dependent diversification. Systematic biology, 65(2), 181-193. https://doi.org/10.1093/sysbio/syv066.

For STRAPP in deep times: Doré, M., Borowiec, M. L., Branstetter, M. G., Camacho, G. P., Fisher, B. L., Longino, J. T., Ward, P. S., Blaimer, B. B., (2025), Evolutionary history of ponerine ants highlights how the timing of dispersal events shapes modern biodiversity, Nature Communications. https://doi.org/10.1038/s41467-025-63709-3

Author

Maël Doré