We conducted a global literature review and meta-analysis to evaluate whether seed size could predict post-dispersal seed predator effects on seed removal and plant recruitment, respectively.  Datasets were built using data extracted from published studies focusing on seed predation by small mammals (see Methods for criteria and data extraction protocol).  We found that seed size predicted small mammal seed removal rates and their impacts on plant recruitment consistent with optimal foraging theory, with intermediate seed sizes most strongly impacted globally - for both native and exotic plants.  However, differences in seed size distributions among ecosystems conditioned seed predation patterns, with relatively larger-seeded species most strongly affected in grasslands (smallest seeds), and relatively smaller-seeded species most strongly affected in tropical forests (largest seeds).  Such size-dependent seed predation has profound implications for coexistence among plants because it may enhance or weaken opposing life-history tradeoffs in an ecosystem-specific manner.  Our results suggest that seed size may serve as a key life-history trait that can integrate consumer effects to improve understandings of plant coexistence. 

We conducted a global literature review to obtain data related to small mammal effects on plant species, as measured in each of two types of studies - each producing a separate dataset included in this archive.  First, we examined studies wherein researchers offered seeds in the field to quantify seed removal rates by post-dispersal seed predators.  Second, we evaluated experiments wherein researchers sowed seeds in paired cages that allowed or precluded seed predator access in order to quantify post-dispersal seed predator effects on plant recruitment.

To obtain studies that quantified small mammal seed removal rates and/or experimentally evaluated the effects of seed removal on plant recruitment, we first considered data from the global literature review conducted by Moles et al. (2003; Ecology, 84, 3148-3161).  We then supplemented these studies with more recent work (published from January 2002 through November 2019) by searching the Web of Science Core Collection for peer-reviewed studies using the search term combination: ("seed remov*" OR "seed predat*" OR "seed offer*") AND ("small mammal*" OR "rodent*" OR "granivor*").  We also screened studies found through this search for relevant citations.

From this literature pool, we filtered studies based on the following criteria.  All studies considered for our evaluation were conducted in natural systems, i.e. not urban or agricultural settings.  In this definition, we included European grasslands, which are often sustained by mowing, and North American old-field systems, which are seral stages of post-agricultural systems returning to their natural state.  We focused on studies examining primarily seeds, not fruits, though studies incorporating seeds removed from fruits were included.  Likewise, seeds attached to well-developed samara or samara-like dispersal structures (e.g., Acer, Fraxinus) were excluded, as these adaptations can affect total diaspore mass and shape in ways that alter seed handling and removal.  Seed offerings above the ground surface were excluded.  Because consumer origins could influence outcomes (Parker et al. 2006), we also excluded studies where small mammal seed predators could be identified as introduced (e.g., from some island systems), as these were too few to address consumer origins in analyses.  In all cases, only studies that reported results by individual plant species could be included.  Additionally, for seed removal studies, given that study-level variables such as study design, seed predator community, and environmental context (e.g., background resource levels, competition, etc.) could have large effects on measured responses, we only included studies that provided data for >1 plant species to control for such variation.  We did not impose this restriction on plant recruitment studies given that such studies were limited in number, but exclusion of single-species studies did not alter meta-analysis results (Appendix S1).

Data from qualifying studies were extracted directly from the text or tables of publications, or from figures using Web Plot Digitizer Version 4.0 (https://automeris.io/WebPlotDigitizer/).  For seed removal studies, we obtained data on the proportion of seeds removed per plant species during seed removal trials.  In studies reporting multiple seed removal rates per species, we used the following rules to select one value (to overcome the complication of trying to account for non-independence among species x study replicates in models).  When seed removal studies presented data for multiple trials or time points per species, we used data from the last or longest reported time interval with one exception: if the removal rate approached 100% for multiple plant species (potentially truncating responses), we used the middle trial or median time point.  In studies presenting data for multiple sites or ecological contexts (e.g., variation by habitat, distance from habitat edge, or forest age), we randomly chose a single scenario.  For studies of seed predator effects on plant recruitment, we extracted data on the mean number of seedlings or proportion of seedlings recruiting per species, associated measures of variation (SD or SE), and sample size for both small mammal exclusion (treatment) and small mammal access plots (control).  In studies reporting small mammal effects multiple times per species, we selected one estimate (for reasons noted above) using the following criteria.  For studies reporting effects over multiple years and/or ecological contexts, we selected the scenario with the highest overall recruitment in the control plots to maximize potential to detect seed predator effects.  The exception to this rule was that in studies including both disturbed and undisturbed treatments, we used data only from disturbed treatments to exclude effects of plant competition, which can vary by seed size. 

Geographic and habitat information was recorded for all studies.  The latter was used to define the following ecosystem types: grasslands (including European grasslands and old-field systems as defined above), temperate forest, tropical forest, and other (e.g., desert, shrublands, coastal dunes).  Plant species provenance (i.e., native or exotic) was taken directly from the study when possible (most cases), and otherwise from the Plants of the World online database (Royal Botanic Gardens Kew 2020).  Seed mass data for studies from Moles et al.’s (2003) global review were obtained from Appendix I in that publication.  For remaining studies, we gathered seed mass data directly from the source publication or a related publication when possible (85% of cases).  If authors did not include seed mass, we obtained this information from the following published databases, prioritized in the order listed: Seed Information Database (Royal Botanic Gardens Kew 2019), LEDA for Northwest European flora (Kleyer et al. 2008), and Botanical Information and Ecology Network Database (Maitner et al. 2018).  In cases where multiple seed masses were given per species in a database, we chose the record from the most proximate geographic region, when given, otherwise we chose the first record listed.  Seed mass (+1) was natural log-transformed for analysis. 

Small mammal seed predation x seed size datasets.

All variables included in the two datasets are detailed on the README tab associated with each file.

Please note that the seed removal dataset was updated to rebut a technical comment by Chen et al. (2021).  The updated seed removal dataset is associated with a separate Dryad archive (https://doi.org/10.5061/dryad.hqbzkh1fp).