Skip to main content

Vocalizations of the squirrel family

Cite this dataset

Newar, Sasha; Bowman, Jeff (2020). Vocalizations of the squirrel family [Dataset]. Dryad.


The dataset Squirrel_Calls is a collection of vocal records (defined as primary literature that numerically describes the vocalization of at least 1 squirrel species) where each row corresponds to a single call type of one species. The details of the row include a summary of the literature metadata, categorical descriptions of the call and the caller as well as numerical values of the call frequencies. The dataset Squirrel_Ecological_Traits is a corresponding set of ecological traits for all the species listed in the Squirrel_Calls dataset. The traits listed (mass, time partitioning, gliding capabilities, habitat, and sociality) reflect hypotheses and predictions explored in the associated article. At the end of this document, there is a complete list of the literature references used to assemble these datasets. Squirrel_Script is the R script used to produce the statistics and models used in the corresponding paper. Squirrel_Tree is a nexus file compiling the data of 1000 trees downloaded from which were subsetted from their published mammalian supertree. The nexus file was used in the R script.


We developed a database beginning with a list of publications that described the vocalizations of squirrels. The minimum requirement for each publication was the description of at least one call with either a spectrographic analysis or numerical data, though the majority of publications described multiple calls per species or described multiple species per publication (493 calls from 72 species represented in 89 publications). The databases used to search for these publications were Google Scholar, JSTOR, Web of Science, and Wiley Online Library. We used the keywords acoustics, acoustic repertoire, calls, frequency, Hz, vocalizations, and ultrasound paired with Sciuridae, squirrel, or an exhaustive list of currently valid and invalid genera (the most updated nomenclature was taken from the Integrated Taxonomic Information System For each call described in the selected publication, the following characteristics were taken: the fundamental frequency (F0: the mean frequency of the primary vibrational frequency of the vocal membrane; kHz), dominant frequency (FDom: the frequency with the greatest energy, power or amplitude; kHz), minimum frequency (FMin: the minimum frequency of the fundamental frequency; kHz), maximum frequencies (FMax: the maximum frequency of the fundamental frequency (or of harmonic on which FDom is measured); kHz), and the highest visible harmonic (FHarm: mean frequency of the highest complete harmonic visible on the spectrograph; kHz).

Once our review of vocalization publications was complete, we searched for the body mass (g), diel activity pattern (diurnal or nocturnal), social complexity, and habitat openness of the dominant habitat (open or closed) of each species from the relevant vocalization papers. If not provided, other resources including Mammalian Species accounts, PanTHERIA (Jones et al. 2009), and the Animal Diversity Web (Myers et al. 2020) were reviewed. Both male and female body masses were initially recorded, but male body size could not be found for Spermophilus taurensis. Male and female body mass were strongly correlated (r = 0.98, p < 0.001), therefore female body mass was chosen to represent squirrel body size. Because we could only assign an adult female body mass to all species, calls that are exclusively produced by males or pups were removed from the dataset before analysis. We pooled all other calls (calls produced by both sexes or females only as well as calls produced by juveniles and adults) as there is little evidence to suggest that juveniles and adults produce acoustically distinct calls across the family (Matrosova et al. 2007, 2011; Schneiderová 2012; Volodina, Matrosova, and Volodin 2010; but see: Nikol’skii 2007). While the initial database included a five-tiered social classification ranging from solitary to colonial (based on the social grades of ground squirrels described by Matějů et al. (2016)), social classes were reduced to social or solitary living to reduce model parameters. Species that exhibit dynamic social structures, such as flying squirrels that engage in social nesting to a greater extent during one portion of the year (Garroway, Bowman, and Wilson 2013), were treated as socially living. Two subspecies (Marmota baibacina centralis and Tamias dorsalis dorsalis) could not be used in the subsequent analyses because ecological data and body mass-specific to each subspecies could not be found; similarly, the species Spermophilus pallidicauda could not be included as body mass for either sex could not be found.


VertLife, an online resource that allows the user to extract pruned trees from vertebrate supertrees, was used to produce 100 pruned trees from the Mammalian supertree (Upham, Esselstyn, and Jetz 2019). Three subspecies had to be incorporated under their parent species, so branch tips were broken in two and subspecies were treated as equivalent to parent species, with branch lengths identical between the parent and subspecies (the addition of a subspecies did not create any polytomies in the tree). Three species are represented by subspecies only: Sciurus aberti kaibensis, Sciurus niger rufiventer, and Callosciurus erythraeus thaiwanensis.

Statistical Analysis

Phylogenetic generalized least square (PGLS) modelling was used to account for the variation in acoustic repertoire that may be explained by phylogenetic relatedness. PGLS models produce a lambda parameter, λ, that represents the degree to which the variance of traits is explained by the phylogenetic relationships in the model. The λ parameter varies between 0 and 1, with 0 representing no phylogenetic trace and 1 representing absolute Brownian motion (Freckleton, Harvey, and Pagel 2002; Martin et al. 2016).

PGLS modelling restricts each species to a single observation (i.e., no subsampling of species permitted). Therefore, the numerous data entries per species had to be reduced. For the fundamental, dominant, maximum, and highest harmonic frequencies, the absolute maximum value for each characteristic reported among all publications was chosen. Likewise, for minimum frequency, the absolute minimum reported frequency was chosen. We use maximum and minimum values rather than the median for a more rigorous test of our hypothesis about method limits.

Body mass and all frequency characteristics were log-transformed to achieve normal distributions. Additive models were built for each frequency type (β0 + body mass (βMass) + diel activity pattern (βDiel) + sociality (βSociality) + habitat openness (βOpen) + method limits (βLim)) using the caper package in R (ver 3.6.2). We reported the test statistics of the regression to evaluate significance and effect size (F-statistic, p-value, and adjusted R2).