Skip to content

Latest commit

 

History

History
160 lines (126 loc) · 6.71 KB

analysis-plan.md

File metadata and controls

160 lines (126 loc) · 6.71 KB
Raw

Analysis plan

This document lists a series of analysis that could be performed for the project manuscript. These analysis are roughly analysis with the aims of the project while acknowledging known data limitations.

These analysis are based on data exploration investigating the sampling bias, and the catch composition.

Species (types) importance

Question: What species (and species-types) are the most important over time in community refuges in Cambodian Rice fisheries?

  • Calculate IRI per refuge, per species, per occasion
  • Construct a multinomial of % IRL per species type. Time of the year is included as a fixed effect. Species and refuge are included as random effects.

Analysis outputs:

  • Figure with mean IRI of species types over the year
  • Table of average species importance (as per IRI) in the Supplementary Information
  • Table with model summaries in the Supplementary Information

Outstanding questions:

  • Does calculating the IRI per refuge/species/occasion vs species/occasions could dramatically affect the results?
  • Should this analysis be simplified to calculate IRI across refuges? This is one IRI per species, per occasion only. This will allow for simpler visualisation of the results as %IRI could be included but we won’t learn much about inter-site variation.
  • Does this analysis overlaps with biodiversity analysis below? It might tell a similar story and using more analysis than needed might complicate the story.

Analysis time: 3-6 days

Biodiversity patterns

What are the differences in biodiversity among sites and across time. What refuge factors influence this diversity?

  • Calculate Shannon’s alpha diversity and evenness per refuge, species, occasion (using base e and Hill’s ratio).
  • Calculate Simpson’s alpha diversity and evenness per refuge, species, occasion (using base 2 and Hill’s ratio again).
  • Further analysis will be performed using Shannon’s diversity indices as it’s slightly more common (and is the one used in PASGEAR). We’ll have Simpson indices ready in case reviewer’s ask for a sensitivity analysis.
  • Construct two models one for diversity and one for evenness. In both models include spatial and temporal factors as covariates. As fixed effects: community category, time of the year (and the interaction with community category). As random effects refuge id, year, and sampling effort.

Analysis outputs:

  • R2 values for both models
  • Graph of conditional responses for each fixed effects variable (or most interesting)
  • Table with models summaries in the Supplementary Information

Outstanding questions:

  • Note that there is a lage overlap between this analysis and Fiorella et al. study. Included here because I saw there still an interest of “investigating Shannon H by sites and by gear type”. However need to be mindful that effort was standardised and gear comparisons would be limited to overlapping times.
  • What are the most important refuge co-variates would be important to add in the model? Those that could be viewed as refuge interventions could be particularly interesting.
  • What sampling covariates could be important when modelling the species diversity? Those that are not easily explained by the time of the year/occasions could be particularly interesting.
  • Note that we are already using a fair number of degrees of freedom and including more than 4 extra factors (degrees of freedom) might be infeasible.

Analysis time: 4-8 days

Community composition (first level beta diversity)

How does community composition changes across time and space (beta diversity measured by Chi2 distances)? Possible extension: Is community composition driven by site characteristics?

  • Correspondence analysis (akin to PCA) of abundance matrix (species/refuges+occasions). Uses Chi-square distances of beta-diversity.
  • Visual representation of first axis to explore/identify patterns across time of the year and refuge type.
  • Concordance analysis to examine whether grouping by time of the year and refuge type is a “significant” way to cluster communities
  • Possible extension step: Canonical correspondence analysis to constrain the ordination. This would answer answers to the additional question of whether composition is driven by site/occasion characteristics. We could include a large number of predictors but the larger the number the more complicated the analysis and interpretation.
  • Possible extension step: Permutation tests to determine the significance of associations between composition and predictors.
  • Possible extension step: Variation partition analysis to determine the relative importance of predictors.

Analysis outputs:

  • Biplot of communities in principal axis space.
  • Possible extension: Triplot of communities and predictors in principal axis space
  • Possible extension: Model outputs in supplementary information

Outstanding questions:

  • Similar as in previous analysis it would be possible to include community info from individual replicates separately or jointly. The full implications of either options are unknown.
  • It is possible that the response of species abundance to site/occasion characteristics is not unimodal. In this case we cannot use the results of the canonical correspondence analysis in the paper, but we won’t know until we try it.

Analysis time: 3-6 days (7-11 days with possible extension)

Temporal beta-diversity index

How similar is the composition between t1 and t2 within a community? Has composition changed in exceptional ways? Is it due to mostly species gains or losses?

This Temporal beta-diersity index TBI has been recently developed by Legendre (2018) Ecology and Evolution

  • Compute TBI
  • Compute signigficancy of differences
  • Identify species that drive significant changes
  • Compute species gains and losses by % denominator

Analysis outputs:

  • Plot of changes in dissimilarity (total and partitioned across gains and losses) across the chosen comparison points in time for each site.
  • Possibly plot of gain/losses across sites and comparison periods

Outstanding questions:

  • Complex analysis as we have more than one time and it’s not clear what comparisons would me more relevant. Should we focus comparison between along years for a single time of the year? Should we focus comparison between seasons and use years as replicates? Should we focus in an start/end of the time series?