Introduction

This R Markdown file is used to bring together all of the results generated from the High Performance Computer (HPC) that will then be saved as combined spreadsheets saved to dropbox. Some of the basic numeric results will be included in this markdown, but all of the data visualization will be left for the 4th and final step of this analysis.

As a reminder, the co-abundance models were implemented on this R script: scripts/HPC_code/HPC_co-abundance_model_final.R. However, the each JAGS model with long MCMC settings is too heavy to download (i.e., several GB per model), so we only saved spreadsheets from the model on the HPC.

Description of results

Total models completed

A total of 258 co-abundance models out of 258 were successfully completed and downloaded from the HPC. The models are composed of 194 ‘community models’ across the entire eligible community, and 64 ‘preferred models’ based on large carnivore dietary preferences. There were a total of 117 community top-down models and 32 preferred top-down models. There were a total of 77 community bottom-up models and 32 preferred bottom-up models.

When examining pairwise comparisons of species, each species-pair (regardless of the direction) can produce one result that is either top-down, bottom-up, or unsupported. There were a total of 16 species-pairs supporting top-down regulation, a total of 29 species-pairs supporting bottom-up regulation, and a total of 98 species-pairs supporting no associations. These results can be further sub-divided into preferred & community results:

  • For the preferred species-pairs, there were 2 species-pairs supporting top-down regulation, a total of 12 species-pairs supporting bottom-up regulation, and a total of 18 species-pairs supporting no associations.

  • For the community species-pairs, there were 14 species-pairs supporting top-down regulation, a total of 17 species-pairs supporting bottom-up regulation, and a total of 80 species-pairs supporting no associations.

The settings used on the HPC to complete these models are composed of 258 models with long settings, 0 models with middle settings, and 0 models with short settings.

If there are any uncompleted preferred models, the names of those species-pairs will appear here: .

Model diagnostics

It is important to inspect the diagnostics of the models before we interpret results, and there are four key diagnostic features to inspect.

  1. The first key value we are interested in is the parameter convergence (known as R-hat) around the species interaction value (SIV), where parameters are converged if values are between 1 and 1.1. The median Rhat for the SIV is 1.001 with a standard deviation of 6.176. In total, 41 models failed to generate a convergent SIV. This is composed of 11 preferred models and 30 community models.

  2. The second key value is the Bayesian p-value denoting model goodness-of-fit, where models are deemed to be a good fit if values are between 0.15 to 0.85, with values outside of this range being considered a bad fit. Across all models 16.67 percent of models are considered a bad fit. This is composed of 11 preferred models and 32 community models.

  3. The third key value is the C-hat overdispersion value, where models are deemed to not have any overdispersion in detections if values are between 0.95 and 1.3. Across all models 100 percent of models contain remaining overdispersion of detections not accounted for in the detection formula. This is composed of 0 preferred models and 0 community models.

  4. The fourth value is the direction of the SIV based on our hypotheses, where top-down models should have a significant negative relationship and bottom-up models should have a significant positive relationship. As a reminder, top-down models are when large carnivores are the dominant species and bottom-up models are where large carnivores are the subordinate species. For the preferred top-down models, 2 models meet these conditions. For the preferred bottom-up models, 14 models meet these conditions. For the community top-down models, 14 models meet these conditions. For the community bottom-up models, 26 models meet these conditions.

Unsupported results

In combining these different diagnostic tools, we can determine different levels of support for our results. Models that had unsuitable parameter convergence or Bayes p-values and remaining over-dispersion are likely due to insufficient data or excessive residual noise from weakly interacting species (Fig. S2, unsupported_wrong_direction). Models that had unclear SIVs are likely due to the inclusion of covariates in the model being more informative than species interactions (Fig. S2, unsupported_unclear_SIV). Models with clear SIVs in the opposite direction as we hypothesized are likely due to shared responses to unexplained variation in covariates not included in the model, such as both predator and prey responding negatively to hunting in bottom-up models, or both species responding positively to fruit availability in top-down models (Fig. S2, unsupported_poor_fit). Finally, when a model meets all of the previous criteria, then we can determine the model as supported. Top-down and bottom-up models may fail for different reasons, so unsupported species-pairs are simply unsupported, while models can be classified into the described levels. The results from each level are provided below:

  1. There were a total of 68 unsupported_poor_fit models. This is composed of 18 preferred models and 50 community models.

  2. There were a total of 90 unsupported_unclear_SIV models. This is composed of 20 preferred models and 70 community models.

  3. There were a total of 55 unsupported_wrong_direction models. This is composed of 12 preferred models and 43 community models.

  4. There were a total of 45 supported models. This is composed of 14 preferred models and 31 community models.

In interpreting our supported results, we can split them between top-down and bottom-up results:

Supported top-down results

  • For the supported top-down results, we had a total of 16 models. This is composed of 2 preferred models and 14 community models.
    • Tigers showed 1 top-down relationships with their preferred prey and 8 top-down relationships with the overall community. In terms of distinct species-pairs, tigers exhibited top-down regulation in 1 preferred species-pairs and 8 in community pairs.
    • Leopards showed 1 top-down relationships with their preferred prey and 2 top-down relationships with the overall community. In terms of distinct species-pairs, leopards exhibited top-down regulation in 1 preferred species-pairs and 2 in community pairs.
    • Dholes showed 0 top-down relationships with their preferred prey and 4 top-down relationships with the overall community. In terms of distinct species-pairs, dholes exhibited top-down regulation in 0 preferred species-pairs and 4 in community pairs.
    • Clouded leopards showed 0 top-down relationships with their preferred prey and 0 top-down relationships with the overall community. In terms of distinct species-pairs, clouded leopards exhibited top-down regulation in 0 preferred species-pairs and 1 in community pairs.

Supported bottom-up results

  • For the supported bottom-up results, we had a total of 29 models. This is composed of 12 preferred models and 17 community models.
    • Tigers showed 2 bottom-up relationships with their preferred prey and 3 bottom-up relationships with the overall community. In terms of distinct species-pairs, tigers exhibited bottom-up regulation in 2 preferred species-pairs and 3 in community pairs.
    • Leopards showed 2 bottom-up relationships with their preferred prey and 4 bottom-up relationships with the overall community. In terms of distinct species-pairs, leopards exhibited bottom-up regulation in 2 preferred species-pairs and 4 in community pairs.
    • Dholes showed 1 bottom-up relationships with their preferred prey and 4 bottom-up relationships with the overall community. In terms of distinct species-pairs, dholes exhibited bottom-up regulation in 1 preferred species-pairs and 4 in community pairs.
    • Clouded leopards showed 7 bottom-up relationships with their preferred prey and 6 bottom-up relationships with the overall community. In terms of distinct species-pairs, clouded leopards exhibited bottom-up regulation in 7 preferred species-pairs and 6 in community pairs.

Specific results per large carnivore

  1. Tigers preferred bottom-up relationships were with the following species: Rusa_unicolor, & Sus_scrofa, and the preferred top-down relationships were with the following species: Muntiacus_genus.
    • The community bottom-up relationships were with the following species: Arctictis_binturong, & Macaca_arctoides, & Paguma_larvata, and the community top-down relationships were with the following species: Atherurus_macrourus & Lophura_nycthemera & Macaca_nemestrina & Manis_javanica & Neofelis_genus & Trichys_fasciculata & Viverra_megaspila & Viverra_zibetha.
  2. Leopards preferred bottom-up relationships were with the following species: Muntiacus_genus, & Sus_scrofa, and the preferred top-down relationships were with the following species: Macaca_nemestrina.
    • The community bottom-up relationships were with the following species: Atherurus_macrourus, & Bos_gaurus, & Rusa_unicolor, & Tapirus_indicus, and the community top-down relationships were with the following species: Catopuma_temminckii & Varanus_salvator.
  3. Dholes preferred bottom-up relationships were with the following species: Rusa_unicolor, and the preferred top-down relationships were with the following species: .
    • The community bottom-up relationships were with the following species: Arctictis_binturong, & Lophura_diardi, & Paradoxurus_hermaphroditus, & Viverra_zibetha, and the community top-down relationships were with the following species: Arctonyx_collaris & Catopuma_temminckii & Helarctos_malayanus & Macaca_fascicularis.
  4. Clouded leopards preferred bottom-up relationships were with the following species: Arctictis_binturong, & Arctonyx_collaris, & Capricornis_genus, & Muntiacus_genus, & Sus_scrofa, & Viverra_tangalunga, & Viverra_zibetha, and the preferred top-down relationships were with the following species: .
    • The community bottom-up relationships were with the following species: Atherurus_macrourus, & Herpestes_brachyurus, & Lophura_bulweri, & Paguma_larvata, & Tapirus_indicus, & Trichys_fasciculata, and the community top-down relationships were with the following species: .

Sensitivity test results

We implemented a total of 5 sensitivity tests that will be compared with the original tests for a total of 6 tests. The key inference here will be determining which species pairs agree across all possible tests, noting that not every species pairing was able to be tested in every way. Currently, all tests have been run with the MCMC setting LONG. It is also important to note that while we ran counter factual tests for all 64 preferred models, we also ran counter factual tests for an additional 31 community models that generated supported results in the original test.

Agreement across sensitivity tests

Finally, we assessed which species pairs agreed across the different tests and this resulted in 2 preferred top-down model and 9 community bottom-up models. This also resulted in 4 community top-down model and 5 community bottom-up models.

  • The 2 preferred top-down models that agree across all sensitivity tests are: SUB-Macaca_nemestrina~DOM-Panthera_pardus, SUB-Muntiacus_genus~DOM-Panthera_tigris.
    • This equates to 2 species pairs composed of the following pairs: Macaca_nemestrina-&-Panthera_pardus, Muntiacus_genus-&-Panthera_tigris
  • The 9 preferred bottom-up models that agree across all sensitivity tests are: SUB-Cuon_alpinus~DOM-Rusa_unicolor, SUB-Neofelis_genus~DOM-Arctictis_binturong, SUB-Neofelis_genus~DOM-Arctonyx_collaris, SUB-Neofelis_genus~DOM-Capricornis_genus, SUB-Neofelis_genus~DOM-Sus_scrofa, SUB-Neofelis_genus~DOM-Viverra_zibetha, SUB-Panthera_pardus~DOM-Muntiacus_genus, SUB-Panthera_tigris~DOM-Rusa_unicolor, SUB-Panthera_tigris~DOM-Sus_scrofa.
    • This equates to 9 species pairs composed of the following pairs: Cuon_alpinus-&-Rusa_unicolor, Arctictis_binturong-&-Neofelis_genus, Arctonyx_collaris-&-Neofelis_genus, Capricornis_genus-&-Neofelis_genus, Neofelis_genus-&-Sus_scrofa, Neofelis_genus-&-Viverra_zibetha, Muntiacus_genus-&-Panthera_pardus, Panthera_tigris-&-Rusa_unicolor, Panthera_tigris-&-Sus_scrofa
  • The 4 community top-down models that agree across all sensitivity tests are: SUB-Catopuma_temminckii~DOM-Panthera_pardus, SUB-Trichys_fasciculata~DOM-Panthera_tigris, SUB-Varanus_salvator~DOM-Panthera_pardus, SUB-Viverra_megaspila~DOM-Panthera_tigris.
    • This equates to 4 species pairs composed of the following pairs: Catopuma_temminckii-&-Panthera_pardus, Panthera_tigris-&-Trichys_fasciculata, Panthera_pardus-&-Varanus_salvator, Panthera_tigris-&-Viverra_megaspila
  • The 5 community bottom-up models that agree across all sensitivity tests are: SUB-Cuon_alpinus~DOM-Arctictis_binturong, SUB-Cuon_alpinus~DOM-Viverra_zibetha, SUB-Neofelis_genus~DOM-Paguma_larvata, SUB-Neofelis_genus~DOM-Tapirus_indicus, SUB-Panthera_tigris~DOM-Arctictis_binturong.
    • This equates to 5 species pairs composed of the following pairs: Arctictis_binturong-&-Cuon_alpinus, Cuon_alpinus-&-Viverra_zibetha, Neofelis_genus-&-Paguma_larvata, Neofelis_genus-&-Tapirus_indicus, Arctictis_binturong-&-Panthera_tigris

Save results

Save these results to GitHub, in this file directory: /results_final/step3_output_combined_results. It is important to note that Tables S3 (full model performance and SIVs) and S4 (supported counterfactual results) are generated from this script, so make sure to include these in the manuscript!