Plant species vary in maximum size across several orders of magnitude. Even within a single community this variation can be great (Schamp and Aarssen 2014) and research suggests it can be ecologically relevant. For example, variation in plant species size appears to contribute to competitive ability (typically competitive effect ability; Gaudet and Keddy 1988; Keddy and Shipley 1989; Goldberg and Landa 1991; Keddy et al. 1994, 2002; Rösch et al. 1997; Keddy 2001; Fraser and Keddy 2005; Violle et al. 2009; Wang et al. 2010; Carmona et al. 2019; but see Gerry and Wilson 1995) and intraspecific density-dependence (self-thinning) scales with species size (White 1980; Enquist et al. 1998; Schamp and Aarssen 2014). Additionally, monocultures of larger species appear to be more susceptible to invasion, particularly by smaller species (Schamp and Aarssen 2010). Finally, evidence suggests that plant species that contribute strongly to ecosystem function tend to be relatively tall (Brun et al. 2022). It is clear that plant species size represents an important functional trait that contributes to a species’ ecological strategy (Westoby 1998). Consequently, it is intriguing that evidence for size-specific community assembly rules is sparse.

A number of ecological studies have examined the impact of variation in plant species maximum size on community assembly. Evidence of size-based community assembly using null modeling approaches, for example, has been inconsistent (Weiher et al. 1998; Stubbs and Wilson 2004; Schamp and Aarssen 2009) or simply lacking (Schamp et al. 2008, 2011). Additionally, few field studies show evidence of any competitive advantage by larger plant species (Aarssen et al. 2014; Schamp et al. 2013; but see Balfour et al. 2022 for evidence of a modest species size advantage).

There is considerable evidence that herbivores are consequential in affecting plant community composition by altering abundance and diversity (Batzli and Pitelka 1971; Howe et al. 2002; Peters 2007; Eskelinen et al. 2022), and additional evidence suggests that herbivore activity, especially by small mammals, can have species-specific effects due to preferential seed or seedling predation (Hulme 1994; Howe and Brown 2000; Howe and Lane 2004; Howe et al. 2006; Sullivan and Howe 2009; Bricker and Maron 2012; Pellish et al. 2018; Dylewski et al. 2020; Steketee et al. 2022). There are several reasons to suspect that herbivory/granivory may be size biased. First, seedlings of smaller species are closer to mature size than seedlings of larger species. Hypothetically, smaller species may deploy anti-herbivore defenses at a smaller absolute size than larger species, leaving the seedlings of larger species relatively less mature and perhaps more susceptible. Second, larger plant species generally have larger seeds (Aarssen and Jordan 2001), which may be more readily found by herbivores and may have an increased nutrient reward per seed (Samson et al. 1992; Grubb and Burslem 1998); large-seeded plant species, particularly in grasslands, appear to be more greatly impacted by granivores (Dylewski et al. 2020). Seed size and shape may also be predictive of persistence in soil, which could reflect the activities of granivores and the ability of small-seeded species to persist (Thompson et al. 1993; Funes et al. 2003; but see Moles et al. 2000). Third, some research suggests that variation in gross morphology, including height/biomass, can be predictive of susceptibility to herbivores, at least for woody plant species (Carmona et al. 2011). A size bias in herbivory is also aligned with predictions from Price’s (1991) plant vigor hypothesis, which suggests that invertebrate herbivores prefer larger leaves, stems, and/or plants and with research suggesting that intraspecific variation in plant size can predict invertebrate herbivore responses (Barbour et al. 2015). Finally, seedling herbivory, regardless of the species eaten, will leave small gaps in the vegetation that can be captured by the dispersules of other species. Some research indicates that smaller species, which have smaller space requirements and can reproduce at a smaller size (Aarssen et al. 2006), are better at colonizing such gaps (Schamp and Aarssen 2010). Consequently, there is ample reason to investigate the possibility that the actions of herbivores influence plant species composition in a size-biased manner.

Several lines of evidence suggest that herbivory is important in our study community, an old field dominated by herbaceous perennial grasses and forbs. A separate seedling transplant experiment indicates a high rate of seedling herbivory (> 75% of 564 transplants across 15 species) in the field (Schamp unpublished). Camera trap data show activity around these transplants by small mammal herbivores, particularly meadow voles (Microtus pennsylvanicus Ord, 1815) and meadow jumping mice (Zapus hudsonius Zimmermann, 1780) (Fig. 1); both of these species are known to regularly consume the seedlings of multiple species found in our community (Hulme 1994; Howe and Brown 2000; Howe and Lane 2004; Howe et al. 2006; Sullivan and Howe 2009; Bricker and Maron 2012; Pellish et al. 2018; Dylewski et al. 2020; Steketee et al. 2022). There is ample reason to suspect that herbivores may be important in driving compositional patterns within this community.

Fig. 1
figure 1

Camera trap image of small mammal herbivore. Based on identification from two colleagues with expertise in small mammals of the region (Albrecht Schulte-Hostedde and Jeff Bowman), the image appears to be a jumping mouse, likely a meadow jumping mouse (circled in white; Zapus hudsonius Zimmermann, 1780)

In this study we investigate whether herbivores/granivores change the composition of an old-field plant community in a way that favors smaller plant species. To accomplish this, we compared the abundance and diversity of large and small species in caged (herbivore exclusion) and control plots. Above, we highlighted several reasons why herbivores/granivores may preferentially eat the seedlings of larger plant species/larger-seeded plant species. Consequently, we have predicted that taller plant species (i.e., species with larger mature sizes) would be underrepresented—and smaller species overrepresented—in control plots compared to herbivore exclusion (caged) plots.


Study site

We performed this study in an old-field plant community at the Ontario Forest Research Institute’s Arboretum in Sault Ste. Marie, Canada (46°32′34.1″ N 84°27′37.4″ W). Vegetation within the field consists of herbaceous, mainly perennial, plant species (Table S1). The field site has remained unmown and untilled since 2007. In 2009, we established 150, 1-m diameter plots within 1.5 m × 1.5 m plot areas randomly located within a 25 m × 67.5 m field grid—plots are separated by 1-m laneways (Fig. S1). Using a random number table, we assigned plots to one of two treatments: control (n = 75) or caged (n = 75). We enclosed caged plots with a 1-cm gauge fence that extends roughly 45 cm above the ground and 15 cm below the ground. We left the tops of cages open to allow for vertical plant growth. The purpose of the caged treatments was to eliminate or at least greatly reduce herbivory. Based on camera trap evidence and general observations, there are a significant number of meadow voles and meadow jumping mice that frequent our study site. While we have previously seen large herbivores (ex. deer) in the arboretum facility (96 ha), we have not witnessed any walking through or feeding within our study community–camera trap data across 40 days from mid-July to mid-September in the summer of 2022 produced no evidence of deer in the study field. It remains possible that deer browse within the community.

Data collection

Beginning nine years after establishing our experimental plots—which allowed for any treatment effects on plant composition to accrue, we recorded the abundance of each plant species in each plot (yearly, from 2018 to 2022). Abundance was recorded as the number of rooted units (i.e., ramets) of each species within each 1 m plot (e.g., Brazeau and Schamp 2019).

To estimate maximum height for each of our species (Westoby 1998), we collected height data continuously throughout the duration of this study, identifying and sampling tall reproductive individuals of each species visible within the bounds of our field site. The median number of values recorded per species is 41 (range = 20−80); we used the largest value per species as the maximum height for that species. Because we were interested in maximum species height as a species-level functional trait representing the potential size each species could reach in our community, and because we did not wish to disturb our experimental plots, all plants measured were in the surrounding non-treatment plots. Our interest was in species-level compositional changes related to potential size, not individual plant-level size in relation to herbivory, so measuring the size of species in individual plots was unnecessary and would not inform our hypotheses. When collecting maximum plant height, we measured the length of the plant from the ground level to the highest point on the plant without manipulation (i.e., we did not stretch the plant while measuring, and in some cases, a plant species reproductive structure was the tallest plant structure). We tested the influence of this choice on our results by testing whether our results changed when we substituted vegetative height data for three rosette species for which the reproductive height is very different from vegetative height (Plantago major, Plantago lanceolata, Pilosella aurantiaca, Details in Table S2).

To determine whether herbivores were active in eating seedlings in our study plots and establish that our caged treatment limits this herbivory, we conducted a transplant experiment. During the summer of 2022, we grew seedlings of Trifolium pratense in a growth chamber and then hardened them with outdoor growth for a period of one week. In August 2022, we transplanted one seedling into each control and caged plot (75 transplants in control plots, 61 transplants in caged plots; 14 caged plots were damaged in 2021, and these plots were removed from 2021 to 2022 census data prior to data analyses). We then monitored these seedlings weekly until October 2022 to determine whether they would be eaten by herbivores.

Data analyses

To determine whether larger and smaller species were overrepresented in caged and control plots, we used our functional trait estimates of maximum species height in conjunction with abundance data from plots to determine the average maximum height of the species found in each plot (abundance weighted) and compared this across treatments. We also compared plot-level species richness and plot-level abundance between caged and control plots over time; all analyses were carried out using R.4.1.3 (R Core Team 2022). We used three approaches when calculating plot-level richness and abundance. We calculated (1) the richness and abundance of all species found growing in plots (total species richness and abundance), (2) the richness and abundance of species in a plot that had maximum heights lower than or equal to the first quartile of species height for all species in the community for a given census year (small species richness and abundance, Table S3; Balfour et al. 2022), and (3) the richness and abundance of species in a plot that had maximum heights greater than or equal to the third quartile of species height for all species in the community for a given census year (large species richness and abundance, Table S3). We fit the response variable and the average height of species in each plot, using a linear mixed-effects model, with treatment as a fixed variable and year and plot as crossed random variables (“lmer” function, “lme4” package v1.1.28; Bates et al. 2015). We used two versions of this model, one with average maximum height of species in plots calculated using reproductive heights and a second version where these plot-level values are calculated using reproductive heights for most species and vegetative heights for three rosette species (P. major, P. lanceolata, and P. aurantiaca). We log transformed these plot-level measures to meet normality assumptions. We fit the response variables, total species richness, small species richness, and large species richness with generalized linear mixed-effects models, using the Poisson distribution and a log-link function, with treatment as a fixed variable and year and plot as crossed random variables (“glmer” function, “lme4” package; Bates et al. 2015). To address overdispersion, we fit the response variables, total abundance, small species abundance, and large species abundance with generalized linear mixed-effects models, using the negative binomial distribution, with treatment as a fixed variable and year and plot as crossed random variables (“glmer.nb” function, “lme4” package; Bates et al. 2015). We checked statistical assumptions by examining the normality of residuals, normality of random effects, linearity (lmer model only), overdispersion and zero-inflation plots (glmer models only), homogeneity of variance, and influential observations plots (“model_check” function, “performance” package v0.10.0; Lüdecke et al. 2021). To evaluate both the statistical and biological significance of our model estimates, we calculated back-transformed estimated marginal means (least-squares means; “emmeans” package v1.7.3; Lenth 2022) with associated 95% confidence intervals.

Using data from our seedling transplant experiment, we compared T. pratense seedling consumption in caged and control plots using a generalized linear model with a binomial distribution (“glm” function, “stats” v.4.1.3). The response variable in this model was the status of each seedling—coded as 1 if the seedling was eaten (all aboveground biomass removed) during the observation period or 0 if the seedling was alive by the end of monitoring—the predictor variable was treatment (caged vs. control). Methods for checking assumptions and reporting significance are described above.


Consistent with our predictions, species found in caged plots were significantly taller compared to in control plots [1.03 m, 95% CI (0.97, 1.10) vs. 0.94 m, 95% CI (0.88, 1.01); F1,146.36 = 4.34, p = 0.037; Fig. 2; Table 1; Fig. S2]. We observed the same significant over-representation of taller species in caged plots when we used vegetative height data for three rosette species (P. major, P. lanceolata, P. aurantiaca; F1,146.42 = 5.43, p = 0.019; Table S2). Total species richness did not significantly differ between caged and control plots (Table S4); however, small species richness was lower in caged plots compared to control plots [0.41, 95% CI (0.29, 0.59) vs. 0.65, 95% CI (0.47, 0.92); x21 = 7.29, p = 0.007; Fig. 2; Table 1; Fig. S3]. Large species richness did not significantly differ between treatments (Table S4). Total abundance was significantly lower in caged plots compared to control plots [223, 95% CI (168, 296) vs. 273, 95% CI (206, 361); x21 = 4.03, p = 0.045; Fig. 2; Table 1; Fig. S4]. Small species abundance was significantly lower in caged plots compared to control plots [0.95, 95% CI (0.46, 1.98) vs. 3.69, 95% CI (1.81, 7.52); x21 = 14.22, p < 0.001; Fig. 2; Table 1; Fig. S5]. Large species abundance did not significantly differ between treatments (Table S4).

Fig. 2
figure 2

Effects of control and caged treatments on mean maximum species height (m; a), small species richness (b), total abundance (c), and small species abundance (d). Squares represent back-transformed estimated marginal means from linear (a) and generalized b, c, d linear models, and whiskers represent 95% confidence intervals. Bars marked with an asterisk indicate significant differences between treatments (*p < 0.05, **p < 0.01, ***p < 0.001). See Table 1 for model details and summary statistics

Table 1 Back-transformed model estimates for the relationship between treatment (control and vs. caged plots) and mean maximum species height (LMM), small species richness (GLMM with Poisson distribution and log-link function), total abundance (GLMM with negative binomial distribution), and small species abundance (GLMM with negative binomial distribution)

Seedlings of Trifolium pratense were eaten significantly more frequently in control plots relative to caged plots (x21 = 56.84, p < 0.001; Table 2). Trifolium pratense seedlings in control plots had an 89% chance of being completely eaten [95% CI (0.80, 0.95); Fig. 3], while seedlings in caged plots had a 13% change of being completely eaten [95% CI (0.07, 0.24); Fig. 3].

Table 2 Back-transformed model estimates for the relationship between treatment (control and vs. caged plots) and herbivore consumption of T. pratense seedlings (GLM with binomial distribution)
Fig. 3
figure 3

Comparison of the proportion of T. pratense seedlings consumption in control vs. caged plots in the 2022 herbivory experiment (control plot n = 75, caged plot n = 61). Gray sections of bars represent seedlings that were eaten during the growing season, and white sections of bars indicate seedlings that were not eaten by the end of the growing season


We examined the potentially size-biased role of herbivores in shaping plant community composition in an old-field plant community. Evidence from our transplant experiment supports the prominence of herbivores within the community, as well as the efficacy of our caged treatment (89% of seedlings in control plots were eaten). Indirect evidence suggests that small mammals are responsible for this seedling herbivory (voles and meadow jumping mice); however, other herbivores, including deer and snowshoe hare, may also play a role. Voles and meadow jumping mice are known herbivores in grasslands, and voles can be particularly impactful herbivores in grassland systems (Howe and Brown 2000; Howe and Lane 2004; Howe et al. 2006); these small mammals appear to be active and consequential herbivores within our study community.

We tested whether small mammal herbivore exclusion produced size-biased impacts on plant species composition in an old-field community using five years of data following an elapsed nine years of experimental treatment. Plant species composition differed significantly across treatments, with smaller plant species modestly but consistently overrepresented within control plots relative to caged plots. Caged plots, where small mammal herbivores were excluded, consistently contained taller species (9 cm taller on average) compared to control plots. Fewer small plant species were consistently found in caged plots, again across the five years of compositional data we examined. These results indicate that small mammal herbivores in our community significantly impact the composition of coexisting plant species, with the apparent effect of modestly benefiting small plant species. Lastly, total abundance and small species abundance were lower in caged plots. This finding expands our understanding of the role that herbivores play in assembling plant communities (Borer et al. 2014) and expands the list of ecological processes that yield changes in composition related to plant species size.

We did not find clear evidence that control plots contained fewer tall plant species or fewer individuals of tall species, as would be expected if herbivores were preferentially targeting the seedlings of these species. While differences in abundance across treatments for some species suggest that some herbivore/granivore activity is species specific in our study system (Table S5), consistent with abundant literature (Howe and Brown 2000; Howe and Lane 2004; Howe et al. 2006; Sullivan and Howe 2009; Bricker and Maron 2012; Pellish et al. 2018; Steketee et al. 2022), these preferences do not appear to be size biased. It is possible that abundances, particularly of large plant species, are simply slow to change across treatments because mature plants are long-lived. However, it would seem reasonable to expect that after 14 years of treatment effects, larger species, if targeted by herbivores (or their seeds targeted by granivores), would have decreased in abundance and/or diversity in control plots. Consequently, direct herbivory or granivory affecting tall plant species does not appear to account for the observed differences in composition among our treatments.

Aside from direct, size-biased herbivory, it is possible that herbivores may favor increased abundance and diversity of small species by preferentially consuming seedlings (Harper 1977; Crawley 1988; Hulme 1996) regardless of what species or size of species is consumed. As mentioned above, there is considerable evidence that herbivores target seedlings in our study community. The seedlings of T. pratense were eaten at a high rate (89% eaten in control plots) and > 75% of 564 seedlings spanning 15 species were eaten in a single growing season in another experiment in this community (Schamp unpublished). These data, combined with the basic observation that in 14 years of study, no single season has come close to enduring a comparable loss of mature plants to herbivores, allow us to conclude with high confidence that herbivores in our study preferentially target seedlings. Frequent seedling predation will necessarily increase the number of small gaps or microsites available and thus increase opportunities for successful colonization (Hoffman et al. 1994). Some evidence suggests that smaller plant species may be more successful in colonizing gaps (Schamp and Aarssen 2010), perhaps due to their ability to survive and reproduce at smaller sizes (Tracey and Aarssen 2014). Therefore, even if seedling consumption is random and not size biased (but see Hulme 1994), seedling herbivory may still favor small species. For example, the observed herbivory on T. pratense, which is a relatively small species, will leave relatively small gaps that plant species with small maximum sizes may still colonize (Schamp and Aarssen 2010). We do not have data to confirm this mechanism, although we are confident that herbivory is benefiting smaller species and that this is not due to herbivores targeting the tall plant species. The mechanism for our results requires further study; for example, other herbivore effects such as those from trampling, burrowing, and changes in litter distribution may also contribute to the observed compositional differences between treatments.


Herbivores play an important role in our study community and do so in a way that consistently favors smaller plant species, allowing them to achieve modest, but consistent higher richness and abundance. Because this effect appears to arise from seedling herbivory across many herbaceous plant species, it is likely to be a stable effect, as seedlings appear to be eaten before anti-herbivore defenses can be deployed. The consistent benefit of this herbivory in favor of small species may reduce or mitigate any competitive advantage enjoyed by larger plant species (Eskelinen et al. 2022), ensuring that natural communities harbor a collection of plant species spanning a large gradient in size.