Inter- and intraspecific competition are usually observed over a few generations but their patterns and consequences are seldom tractable in natural systems over longer timescales relevant to macroevolutionary change. Here, we use win-draw-lose competitive overgrowths for a marine benthic community of encrusting bryozoans that have evolved together in New Zealand for at least 2.3 million years to investigate battles for substrate space, a resource that is limiting for these colonial organisms. Using more than 6000 fossilized and contemporary battles, we explored what combination of traits – including relative zooid (module) size, weapons, armor and relative abundance –best predict battle outcomes, and if these are time-varying. In simpler models where we disregard trait-trait interactions, we find that the effects of larger zooid sizes and three-dimensional growth on battle outcomes are both positive, while that of relative abundance is negative, such that more common species are more often overgrown by less common species. When we include trait-trait and trait-time interactions in our models, we confirm that a larger zooid size is advantageous for successful overgrowth but infer that it is time-varying. In these complex models, we also detect interactions between combinations of traits, where more armored and weaponized bryozoans seem to be at a disadvantage in overgrowth battles. Our best models do perform statistically better than the null model, but we find that there is low predictability for overgrowth outcomes for our New Zealand dataset of fossil and contemporary battles, suggesting unmeasured variables and/or high stochasticity in a system that is otherwise well-characterized. A best model developed using our New Zealand data is applied to three ecologically similar systems described in previous studies to investigate its general predictive powers, with the expectation that it would perform better than null models. Surprisingly, we find that the best model developed within the New Zealand system cannot be extrapolated to other encrusting cheilostome bryozoan communities and that these three communities often even have opposite signs for trait coefficients. We conclude that there is much to learn about multi-species marine communities where biotic interactions such as competition may have long-lasting consequences for ecological and evolutionary dynamics.
battles
Contains pairwise bryozoan colony battle outcomes, assessments of whether one or the other combatant was larger and difference in relative commonness. A few other trait differences are also given, but these are recalculated in the production line, which also adds other traits and trait-trait interaction terms.
species_traits
Contains trait values for each species found in the battle outcome file.
FormationAges
Contains a description of each site and the age it represents. Used for assigning age numbers in the trait-age interaction terms.
winlose_production_line_bic
The production line scripts were made in order to turn a file describing colony battle outcomes and the species of the two colonies into a response, explanation variable file fit for regression analysis. The explanation variables are trait differences (thus ensuring reshuffling symmetry) made using a separate trait file describing the traits of each species. Interaction terms are made in addtion, before a random subset of battles are reshuffled, meaning that the outcome and each explanation variable is reversed. This in order to make sure there are both wins and losses in the output file, which then can be used in regression. This production line was used for making a single regression file, usable for for instance (but not restricted to) BIC-based model search on regression models (as opposed to a training-validation-test-based model search).
winlose_production_line_tvt
See the description for winlose_production_line_bic for details concerning turning a battle outcome and species trait file into a reshuffled regressison file. The training-validation-test (tvt) version of this production line separates the species into a training set (battles are in the training set if both combatants are in the training species set), a validation set (battles are in the validation set if at least one combatant is in the validation species set and none are in the test species set) and a test set (battles are in the test set if at least one combatant is in the test species set). Instead of one regression files, three are thus produced.
winlose_analyze_bic
This collection of files contains the regression data (the result of running the production line scripts) and the analysis scripts for BIC-based model search. It contains one analysis script for the set of simple explanation variables and another for the full explanation variable set (including interaction terms and using random search). It also contains similar analysis scripts for the 3 external datasets.
winlose_analyze_tvt
This collection of files contains the training, validation and test regression data files (the result of running the production line scripts) and the analysis scripts for training-validation-test-based model search. It contains one analysis script for the set of simple explanation variables and another for the full explanation variable set (including interaction terms and using random search).
winlose_present
This contains an R script (and accompanying regression datasets) for presenting the result of the BIC-based model selection search, including regression coefficient estimates for the best model, variance decomposition, prediction strength assessment and tests on external datasets.
winlose_reselection
This set of scripts resamples the data in various ways in order to answer whether the results of the BIC model selection could come about due to pure chance, random species effects or random species-species pair effects.