Predicting body mass in Ruminantia using postcranial measurements
Data files
Dec 27, 2024 version files 703.08 KB
-
akaike_weights_all.csv
5.47 KB
-
all_ppe_species.csv
68.21 KB
-
body_mass_data.csv
39.66 KB
-
coef_info_by_bone.csv
6.68 KB
-
fossil_data.csv
1.11 KB
-
fossil_merge_groups.csv
356 B
-
MCC_tree.tree
338.77 KB
-
model_info_by_bone.csv
12.53 KB
-
predict_body_mass.Rmd
57.45 KB
-
raw_postcranial_measurements.xlsx
168.66 KB
-
README.md
4.18 KB
Abstract
Size plays an important role in mammalian ecology. Accurate prediction of body mass is therefore critical for inferring aspects of ecology in extinct mammals. The unique digestive physiology of extant ruminant artiodactyls, in particular, is suggested to place constraints on their body mass depending on the type of food resources available. Therefore, reliable body mass estimates could provide insight into habitat preferences of extinct ruminants. While most regression equations proposed thus far have used craniodental predictors, which for ungulates may produce misleading estimates based on indirect relationships between tooth dimensions and size, post-cranial bones support the body and may be more accurate predictors of body mass. Here, I use phylogenetically informed bivariate and multiple regression techniques to establish predictive equations for body mass in 101 species of extant ruminant artiodactyls based on 56 post-cranial measurements. Within limb elements, stepwise multiple regression models were typically preferred, though bivariate models often received comparable support based on AIC scores. The globally preferred model for predicting mass is a model including both proximal and distal width of the humerus, though several models from the radioulna received comparable support. In general, widths of long bones were good predictors, while lengths and midshaft circumferences were not. Finally, I show that where the best elements for prediction are unavailable for fossil taxa, selection of the model with lowest perecent prediction error for the lowest level clade to which the fossil can be assigned could be a productive and novel way forward for predicting mass and subsequently aspects of ecology in fossil mammals.
README
This document contains descriptions of each of the items involved with analyses for "Predicting body mass in Ruminantia using post-cranial measurements"
- "akaike_weights_all.csv": This table is generated from the R.markdown file "predict_body_mass.Rmd" and contains AIC (Akaike's information criterion) values for all models compared to each other. Three values can be found: fit (the information score or AIC score for the model), delta (the difference between each model and the optimal model), and w (the Akaike weight of the model).
- "all_ppe_species.csv": This table is generated from the R.markdown file "predict_body_mass.Rmd". It gives the percent prediction error (PPE or %PE) for each individual species by each regression model generated in this study to predict body mass. PPE is calculated using the equation: [(Observed Mass - Predicted Mass)/Predicted Mass] * 100.
- "body_mass_data.csv": This excel file has the average body mass (kg) for each species and species-averaged postcranial linear measurements for each species. Family, subfamily, and tribe level information are also included. This is the file used for analyses in the R.markdown file "predict_body_mass.Rmd".
- "coef_info_by_bone.csv": This table is generated from the R.markdown file "predict_body_mass.Rmd". It lists the model, which bony element it belongs to (Group), the slope, intercept and residual standard error (RSE) for each regression model.
- "fossil_data.csv": This excel file has the linear measurement data for three fossil specimens: Aletomeryx sp. (F:AM 42883), Cosoryx furcatus (AMNH 51031), and Bison antiquus (AMNH 130792). Empty cells are intentional and mean that the measurement was not available for that specimen. This file is used to generate body mass estimates in the "predict_body_mass.Rmd" script.
- "fossil_merge_groups.csv": This table is generated from the R.markdown file "predict_body_mass.Rmd" and contains the combined body mass estimates for each fossil based on the regressions created from extant data. Cells with "NA" are intentional and reflect missing measurement(s) that were not available for that specimen, so mass was unable to be estimated using that particular regression.
- "MCC_tree.tree": This is a maximum clade credibility tree from Zurano et al. (2019). This time-scaled molecular phylogeny is derived from Bayesian inference of topology branch lengths using mitochondrial genomes for 206 living and recently extinct ruminant species and 21 fossil calibrations. It is read into "predict_body_mass.Rmd" for all phylogenetic analyses.
- "model_info_by_bone.csv": This table is generated from the R.markdown file "predict_body_mass.Rmd". It details for each model which bony element the model is associated with (Group), the sample size (N), the AIC score (AIC), delta AIC (delta), and the Akaike weight (AICw). It also gives a lambda value (lambda), residual standard error (RSE), and log mean (Log.Mean.PPE), minimum (Log.Min.PPE), and maximum percent prediction error (Log.Max.PPE) for each regression model.
- "predict_body_mass.Rmd": This is an R.markdown file containing the scripts for all analyses including bivariate regressions, stepwise multiple regressions, and calculation of percent prediction error for all regressions. First, bivariate regressions are calculated for each individual measurement taken across the postcranial skeleton. Then, stepwise multiple regressions are calculated using all measurements for each individual bone. Finally, stepwise multiple regressions are conducted with only measurements pertaining to proximal or distal regions of some long bones. This script also has the analyses for predicting body mass for the three fossil specimens named in the manuscript.
- "raw_postcranial_measurements.csv": This .csv file contains all of the raw measurement data for each individual used to create a species average for this study. Empty cells are intentional and mean that the measurement was not available for that particular specimen. Images of each postcranial measurement can be found in Figure 2 of the manuscript. Descriptions of each measurement are found in Table S3 of the Supporting Information.
Methods
This dataset was collected by taking linear measurements on postcranial bones in museum collections. The data has been processed using R scripts.