Conspecifics, not heterospecifics, limit individual body condition in a desert rodent community

Abstract

Modern coexistence theory makes an excess of intraspecific over interspecific density dependence the necessary condition for stabilisation, yet the benchmark quantification is entirely botanical. We estimated per-capita interaction coefficients on an individual fitness component, size-adjusted body mass, for the twelve abundant rodent species of the Portal Project record (30,320 measured individuals, 24 plots, 1977-2002). Eleven of twelve species were limited more by conspecifics than by heterospecifics (sign test p = 0.0032), and none showed significant interspecific competition. Substituting one heterospecific neighbour for one conspecific at constant total captures cost 2.07% of body mass (bootstrap 95% CI -2.49 to -1.61), but this unweighted species mean is pulled by two sparse species: seven of twelve contrasts are weaker than -2%. Restricting to structurally adult animals roughly halved it, to -1.119% (-1.676 to -0.644). Size-similar heterospecifics were not detectably stronger competitors. The ordering, not the magnitude, is robust.

Keywords: coexistence theory; competition coefficients; body condition; density dependence; desert rodents; limiting similarity; long-term monitoring; Portal Project; self-limitation; stabilising mechanisms.

Introduction

Modern coexistence theory reduces the maintenance of diversity to stabilising mechanisms, which make each species limit itself more than it limits its competitors, and equalising mechanisms, which shrink average fitness differences (Chesson 2000). The stabilising ingredient is a measurable inequality between per-capita interaction coefficients, and its magnitude is therefore one of the central empirical quantities in community ecology. Adler et al. (2018) supplied the benchmark: across 39 terrestrial plant studies, where both effects were competitive, self-limitation was four to five-fold the stronger, and in almost all remaining pairs intraspecific competition combined with interspecific facilitation, which is also stabilising.

That synthesis is botanical, and the plant ratio need not carry over to mobile vertebrates, which add interference competition and behavioural dominance to exploitative resource use. Desert granivorous rodents are the obvious test system, and at Portal, Arizona, the evidence for interspecific competition is unusually strong: excluding kangaroo rats (Dipodomys spp.) from fenced plots raised the abundance of five smaller granivores over 13 years (Heske et al. 1994). But an abundance response to deleting a whole guild member, with its indirect effects on vegetation and seed banks, is not the per-capita coefficient theory specifies. Eccard & Ylonen (2003) argued exactly this for small mammals: population-level removals miss the individual currencies through which rodent competition operates, and the two levels can disagree.

We therefore measured interaction coefficients on an individual fitness component. Body mass standardised for structural size is the conventional condition index in field mammalogy (Peig & Green 2009), and at Portal it is recorded for tens of thousands of handled animals alongside every other animal caught in the same plot on the same nights, yielding a direct estimate of what one extra neighbour of each kind costs an individual.

We tested two predictions, stated before the analysis was run.

H1 (stabilisation). For each focal species $s$, the per-capita effect of a conspecific neighbour on size-adjusted body mass is more negative than that of a heterospecific neighbour, $\beta^{\text{intra}}_s < \beta^{\text{inter}}_s$, and the community mean substitution contrast $\Delta$ is negative with a bootstrap 95% interval excluding zero. H1 is refuted if $\Delta \ge 0$ or if its interval spans zero.

H2 (limiting similarity). Among heterospecific neighbours, those within a factor of two of the focal species in mean body mass depress condition more than those differing by more than a factor of two, $\beta^{\text{sim}} < \beta^{\text{dis}}$. This follows Bowers & Brown (1982), who found that desert granivores with mass ratios below about 1.5 co-occur less often than chance. H2 is refuted if $\beta^{\text{sim}} \ge \beta^{\text{dis}}$ or if the interval on the difference spans zero.

We also examined the prediction of Yenni et al. (2012) that self-limitation is strongest in the rarer members of a community. Following Broekman et al. (2019), we read excess self-limitation on a fitness component as evidence that an interaction is stabilised, not as a demonstration of stable coexistence; because single-component studies give smaller intra-minus-inter differences than full-life-cycle studies (Adler et al. 2018), ours is a conservative test.

Material and Methods

We analysed the Portal Project Teaching Database, the long-term rodent monitoring record of a Chihuahuan Desert site near Portal, Arizona, distributed in the R package ratdat and mirrored by the Rdatasets archive (Ernest et al. 2018; https://vincentarelbundock.github.io/Rdatasets/csv/ratdat/surveys.csv, CC0 1.0), retrieved on 8 September 2026. The file holds 35,549 rows, one per captured individual, from 24 fenced plots between 1977 and 2002; the peer-reviewed data papers for the full research database are Ernest et al. (2009, 2016). We used species_id (two-letter code), plot_id (1-24), year and month (the trapping session), sex, hindfoot_length (mm, the standard structural-size measure), weight (g) and the parsed capture date. Measurements are incomplete as distributed: sex is missing for 2,511 records, hindfoot length for 4,111 and weight for 3,266. Sample construction and the resulting counts are in Table 1.

Model Description

Let $i$ index a captured individual of focal species $s$, caught on plot $p$ in trapping session $t$ (a calendar year-month). Write $W_i$ for body mass (g), $L_i$ for hindfoot length (mm), $C_i$ for the number of conspecifics other than $i$ caught on plot $p$ in session $t$, $H_i$ for the number of heterospecific community members caught in the same plot-session, and $D_i$ for the number of distinct trapping nights in that plot-session. For each species separately we fit

by ordinary least squares, with $\eta$, $\theta$ and $\kappa$ fixed effects for plot, calendar year and calendar month. Entering $\log L_i$ alongside the density terms is a one-stage analysis of covariance rather than a two-stage scaled mass index (Peig & Green 2009), because the weak within-species mass-length correlation makes a standardised major axis slope unstable. Coefficients are multiplied by 100 and read as per cent change in size-adjusted body mass per additional neighbour, a currency comparable across species differing twenty-fold in mass.

Because $C_i + H_i + 1 = T_i$, the total community captures in that plot-session, equation (1) is an exact reparameterisation of a model containing total captures:

so that the quantity

is the effect of substituting one heterospecific neighbour for one conspecific at constant total captures. Any artefact acting through the total number of animals caught, trap saturation or session-level effort, is absorbed by the $T_i$ term and cannot generate $\Delta_s$. This equivalence is exact only because neighbour effects enter equation (1) linearly in counts: if the response to crowding saturates, the linear $T_i$ term absorbs it only approximately and the protection is correspondingly partial. $\Delta$ is the pre-specified test statistic for H1.

Two further density constructions were fitted. To express neighbours in resource-use rather than head-count units we rescaled heterospecific abundance into conspecific metabolic equivalents by Kleiber scaling,

with $\bar M_j$ the mean body mass of species $j$ in grams. To test H2 we split heterospecific abundance by body-size distance,

and estimated $\beta^{\text{sim}}_s$ and $\beta^{\text{dis}}_s$ jointly with $\beta^{\text{intra}}_s$. Table 2 lists every quantity, its units and its biological justification.

Table 1. Data provenance and analysis sample.

Notes. Rows are captures, not site-visit summaries, so neighbour abundance is an index rather than a density. The community filter excludes unidentified codes and taxa with no recorded mass, including the non-rodent codes present in the file. All counts are printed by the analysis script from the file as loaded.

Simulation Protocol

This subsection describes the analysis sample, the replicate structure and the resampling protocol; there is no simulation, and every quantity is estimated from the real capture records described above. The experimental unit at Portal is the plot, so uncertainty in each community-level quantity comes from a cluster bootstrap that draws all 24 plots with replacement and refits equations (1), (4) and (5) for all 12 species, recomputing the community means each time. We used 1,000 resamples with seed 20260908; the bootstrap is the only stochastic element, so the run is deterministic given that seed. Resampling whole plots makes the intervals robust to arbitrary within-plot dependence, including repeated captures of the same untagged individual. Species-level coefficients carry cluster-robust standard errors with plot-sessions as clusters.

Two severity checks, specified in advance, were executed on the same protocol. The metabolic check replaces $H_i$ by $\tilde H_i$ from equation (4). The adult check keeps only individuals at or above their species' median hindfoot length, because reproductive pulses put many small young animals into the traps at once and could mimic intraspecific competition; it removes structurally immature animals without an age variable, which the record does not contain.

Table 2. Variables and parameters, units and biological justification.

Notes. Species mean masses are computed from the same file. Fixed effects for plot, year, month and sex are nuisance parameters and are not reported.

Analysis

For each species we report $\beta^{\text{intra}}_s$, $\beta^{\text{inter}}_s$ and $\Delta_s$ with cluster-robust standard errors and two-sided $p$-values. Community-level quantities are unweighted means across the 12 species, each species a replicate, with percentile bootstrap 95% intervals and the bootstrap standard deviation; because the mean is unweighted, we also report the full spread of species-level estimates and rank-based tests of the ordering. We counted the species showing the predicted ordering, tested that count with an exact one-sided sign test, and compared the coefficient sets with a paired $t$-test and a Wilcoxon signed-rank test, the latter insensitive to extreme values. Where both coefficients were negative we computed the ratio $\beta^{\text{intra}}_s / \beta^{\text{inter}}_s$, the quantity comparable to the plant benchmark. The Yenni et al. (2012) prediction was tested by correlating $\beta^{\text{intra}}_s$ with $\log_{10}$ total captures.

Results

H1: self-limitation exceeds interspecific limitation

The analysis ran to completion on 30,320 individuals; species-level estimates are in Table 3. Eleven of the 12 species have a negative conspecific coefficient, ten of them individually significant with cluster-robust errors, and not one species has a significantly negative heterospecific coefficient. The single exception is Perognathus flavus, the smallest species at 7.92 g, whose conspecific coefficient is positive and non-significant (+0.379%, $p$ = 0.219).

Table 3. Per-capita neighbour effects on size-adjusted body mass, by species.

Notes. Coefficients are per cent change in size-adjusted body mass per additional neighbour caught in the same plot and trapping session, from equation (1); standard errors are cluster-robust with plot-sessions as clusters. $n$ is the number of individuals with both weight and hindfoot length. mean $C$ and mean $H$ are the average conspecific and heterospecific neighbour counts experienced by the analysed individuals. $\Delta_s = \beta^{\text{intra}}_s - \beta^{\text{inter}}_s$ is the substitution effect at constant total captures. Rows are ordered by $n$. Two-sided $p$-values below 0.0001 are shown as <0.0001. Median model $R^2$ across species is 0.227. Species-level contrasts span +0.307% to -7.360%; the two steepest, Neotoma albigula and Peromyscus maniculatus, come from species averaging fewer than 1.1 conspecifics per plot-session and are the least well conditioned in the set.

At the community level (Table 4), one additional conspecific is associated with a 2.009% reduction in size-adjusted body mass (95% CI -2.456 to -1.547; bootstrap SD 0.233), one additional heterospecific with +0.061% (95% CI -0.048 to +0.152; bootstrap SD 0.051), indistinguishable from zero. The substitution contrast is $\Delta = -2.070$% per individual (95% CI -2.489 to -1.611; bootstrap SD 0.229). H1 is supported: the interval excludes zero by a wide margin.

That magnitude should be read with care. $\Delta$ is an unweighted mean over species, and Table 3 shows contrasts spanning +0.307% to -7.360%. Two sparse species, Neotoma albigula (-7.360%) and Peromyscus maniculatus (-6.512%), lie far outside the rest; the other ten fall between +0.307% and -2.682%, and seven of the twelve are weaker than -2%. The mean is therefore pulled well below the typical species, and -2.070% should not be read as the representative per-capita cost. The ordering, not the point magnitude, is what is robust (Table 5). Eleven of 12 show $\beta^{\text{intra}}_s < \beta^{\text{inter}}_s$ (exact sign test $p$ = 0.0032; Wilcoxon signed-rank $p$ = 0.0024, a rank test insensitive to the two extreme values; paired $t$ = -2.905, $p$ = 0.0143). In the four species where both coefficients are negative, self-limitation is a median 8.53-fold stronger (mean 11.22, range 2.00-25.83). In seven further species intraspecific competition pairs with a weakly positive heterospecific coefficient, the combination Adler et al. (2018) also found to dominate their non-competitive pairs.

Table 4. Community-level per-capita effects and severity checks.

Notes. Unweighted means across the 12 species, each species a replicate; brackets are percentile 95% intervals from 1,000 cluster bootstrap resamples of the 24 plots (seed 20260908). Bootstrap standard deviations for the primary row are 0.233 (intraspecific), 0.051 (interspecific) and 0.229 (contrast). Units are per cent change in size-adjusted body mass per additional neighbour; in the metabolic row the interspecific unit is per conspecific metabolic equivalent. The adult subsample retains individuals at or above the species median hindfoot length. Because the means are unweighted, they are pulled by the two species with the steepest coefficients and should not be read as typical per-capita effects; the species-level spread is in Table 3.

Both severity checks leave the ordering intact. Weighting heterospecific neighbours by resource use rather than counting them alike, so that a 160 g woodrat and an 8 g pocket mouse are no longer equivalent, changes $\Delta$ only from -2.070% to -2.072% (95% CI -2.513 to -1.581). Restricting to structurally adult individuals roughly halves the effect, to $\Delta = -1.119$% (95% CI -1.676 to -0.644), the mean conspecific coefficient falling from -2.009% to -1.098%; recruitment pulses therefore account for part of the raw intraspecific coefficient but not for the ordering, which stays clear of zero.

One qualification applies to every interspecific estimate reported here. $\beta^{\text{inter}}_s$ is a single pooled coefficient over all eleven heterospecific community members, so what we estimate is the absence of a detectable net interspecific effect: pairwise coefficients of opposite sign could cancel within that pool. We did not estimate pairwise coefficients, and the size-similarity split below is the only disaggregation we fitted.

H2: no detectable effect of body-size similarity

H2 is not supported. Splitting heterospecific neighbours at a factor-of-two mass ratio gives $\beta^{\text{sim}} = -0.181$% (95% CI -0.482 to +0.126) and $\beta^{\text{dis}} = +0.127$% (95% CI +0.022 to +0.216). The difference is in the predicted direction, -0.308%, but its interval, -0.614 to +0.026, contains zero. Body-size distance therefore does not measurably modulate interspecific effects on condition here; because the test rests on 12 species, several with no size-similar neighbour, we read this as inconclusive rather than as a refutation of limiting similarity.

Self-limitation and rarity

Conspecific coefficients scale with commonness in the direction Yenni et al. (2012) predicted: $\beta^{\text{intra}}_s$ becomes less negative as total captures rise (Pearson $r$ = 0.613, $p$ = 0.034; Spearman $\rho$ = 0.720, $p$ = 0.008). The two most extreme coefficients, Neotoma albigula at -6.97% and Peromyscus maniculatus at -6.28%, belong to species averaging fewer than 1.1 conspecifics per plot-session, so rarity and weak conditioning are confounded in this correlation.

Table 5. Ancillary statistics across the 12 species.

Notes. Significance of species-level coefficients is at $\alpha$ = 0.05 with cluster-robust standard errors. The ratio is computed only where both coefficients are negative, the subset comparable to the plant benchmark of Adler et al. (2018).

Discussion

Across 26 years and 24 plots, individual rodents at Portal pay a measurable cost for conspecific neighbours and no detectable net cost for heterospecific ones. Eleven of twelve species show that ordering, and where both coefficients are competitive self-limitation is a median 8.5-fold stronger, at or above the four to five-fold ratio Adler et al. (2018) report for plants; in most species interspecific effects are weakly positive, the same intraspecific-competition-with-interspecific-facilitation combination they found. The stabilising ingredient of coexistence theory (Chesson 2000) is thus no weaker in this mobile vertebrate community than in sessile plants, at least in the currency of individual condition. It is the ordering we emphasise, not the mean magnitude, which two sparse species inflate.

This does not contradict the exclosure record. Heske et al. (1994) showed that removing kangaroo rats raises the abundance of smaller granivores, and our result is silent about that: abundance and condition are different currencies on different timescales, and an interspecific effect undetectable within a trapping session may still accumulate into a population response over years through vegetation and seed-bank pathways. What our estimates add is that the instantaneous per-capita interaction an individual experiences is overwhelmingly with its own species, the divergence between population- and individual-level readings that Eccard & Ylonen (2003) anticipated. Because the interspecific coefficient pools all eleven heterospecifics, this is a statement about the net neighbourhood, not about every pair within it.

The interpretation must stop short of coexistence. Following Broekman et al. (2019), excess self-limitation on a fitness component shows that an interaction is stabilised; it is neither necessary nor sufficient for stable coexistence, which also requires the average-fitness-difference condition our data cannot address. Because single-component studies understate the intra-minus-inter difference (Adler et al. 2018), our estimate is a lower bound. That conspecific coefficients are steepest in the rarest species ($\rho$ = 0.720) fits the mechanism Yenni et al. (2012) proposed for the persistence of rare species, but with 12 species, and with rarity confounded with estimation precision, the correlation is suggestive rather than decisive.

The failure of H2 is informative. Bowers & Brown (1982) inferred limiting similarity from co-occurrence: granivores of similar mass are found together less often than chance. We asked whether similar-sized neighbours impose a stronger per-capita cost where they do co-occur, and the answer is unresolved. Co-occurrence and interaction strength may decouple, because the pairs whose interactions would be strongest are the ones assembly has already filtered out; or the test may simply lack power. We do not choose between these here.

Several limitations bound these conclusions. Neighbour abundance is captures per plot per trapping session, an effort-dependent index rather than a density, so the absolute coefficients carry no density units; only the substitution contrast, which by equation (2) holds total captures constant, is protected from session-level effort and trap saturation, and it is the contrast we use for inference. The intraspecific coefficient is partly compositional: reproductive pulses deliver many small young animals at once, and the adult restriction halves the effect, so some of the raw self-limitation is demographic rather than competitive. A second compositional pathway we cannot close is reproductive condition. Mass at fixed structural size includes reproductive burden and gut contents, pregnancy and lactation raise female mass appreciably, and reproduction in this system is itself density- and rainfall-dependent; the dataset records no reproductive status, so this pathway is uncontrolled and could push the conspecific coefficient in either direction. Relatedly, the exact reparameterisation in equation (2) assumes neighbour effects are linear in counts; we fitted no saturating or log-density specification, so nonlinear crowding responses are not excluded and the contrast's protection against effort artefacts is correspondingly incomplete. Seasonality and interannual variation enter additively as month and year fixed effects, whereas a session-level (year-by-month) fixed effect would control rainfall-pulse timing more strictly; that specification was not fitted either. The two steepest coefficients come from species averaging fewer than 1.1 conspecifics per plot-session, are the least well conditioned in the set and inflate the community mean, and one species runs against the pattern altogether. The teaching subset carries no plot treatment labels, so the kangaroo-rat exclosure design is absorbed by plot fixed effects rather than modelled, and the interspecific coefficient averages over plots on which some heterospecifics were experimentally excluded (Ernest et al. 2016). It carries no individual tags, so repeated captures of one animal cannot be identified; the plot-level cluster bootstrap accommodates that dependence but cannot remove it. Finally, this is one site and one 26-year window over which the community's compensatory capacity demonstrably changed (Diaz & Ernest 2022), so our coefficients are averages over regimes, not constants.

The practical implication is that individual condition offers a cheap, high-resolution readout of density dependence from data mark-recapture programmes already collect: whether a declining population is limited by its own numbers or by a putative competitor may be visible in body mass within a single trapping session, long before it appears in abundance.

Acknowledgements

We thank the Portal Project and its field workers for three decades of open, meticulously documented monitoring, and the maintainers of the ratdat package and the Rdatasets archive for distributing the teaching database.

Author Contributions

The study was conceived, designed, analysed and written by the author. The analysis code was written and executed by the author; all numbers reported come from that single executed run.

Data Availability Statement

No new data were generated. The analysis uses the Portal Project Teaching Database, distributed under a CC0 1.0 public domain dedication in the R package ratdat and mirrored at https://vincentarelbundock.github.io/Rdatasets/csv/ratdat/surveys.csv (35,549 rows; retrieved 8 September 2026). The dataset should be cited as Ernest et al. (2018); the peer-reviewed data papers for the full research database are Ernest et al. (2009, 2016).

References

Adler, P.B., Smull, D., Beard, K.H., Choi, R.T., Furniss, T., Kulmatiski, A., Meiners, J.M., Tredennick, A.T. & Veblen, K.E. (2018) Competition and coexistence in plant communities: intraspecific competition is stronger than interspecific competition. Ecol. Lett., 21, 1319-1329.

Bowers, M.A. & Brown, J.H. (1982) Body size and coexistence in desert rodents: chance or community structure? Ecology, 63, 391-400.

Broekman, M.J.E., Muller-Landau, H.C., Visser, M.D., Jongejans, E., Wright, S.J. & de Kroon, H. (2019) Signs of stabilisation and stable coexistence. Ecol. Lett., 22, 1957-1975.

Chesson, P. (2000) Mechanisms of maintenance of species diversity. Annu. Rev. Ecol. Syst., 31, 343-366.

Diaz, R.M. & Ernest, S.K.M. (2022) Maintenance of community function through compensation breaks down over time in a desert rodent community. Ecology, 103, e3709.

Eccard, J.A. & Ylonen, H. (2003) Interspecific competition in small rodents: from populations to individuals. Evol. Ecol., 17, 423-440.

Ernest, S.K.M., Valone, T.J. & Brown, J.H. (2009) Long-term monitoring and experimental manipulation of a Chihuahuan Desert ecosystem near Portal, Arizona, USA. Ecology, 90, 1708.

Ernest, S.K.M., Yenni, G.M., Allington, G., Christensen, E.M., Geluso, K., Goheen, J.R., Schutzenhofer, M.R., Supp, S.R., Thibault, K.M., Brown, J.H. & Valone, T.J. (2016) Long-term monitoring and experimental manipulation of a Chihuahuan desert ecosystem near Portal, Arizona (1977-2013). Ecology, 97, 1082.

Ernest, M., Brown, J., Valone, T. & White, E.P. (2018) Portal Project Teaching Database. figshare. DOI: 10.6084/m9.figshare.1314459.

Heske, E.J., Brown, J.H. & Mistry, S. (1994) Long-term experimental study of a Chihuahuan Desert rodent community: 13 years of competition. Ecology, 75, 438-445.

Peig, J. & Green, A.J. (2009) New perspectives for estimating body condition from mass/length data: the scaled mass index as an alternative method. Oikos, 118, 1883-1891.

Yenni, G., Adler, P.B. & Ernest, S.K.M. (2012) Strong self-limitation promotes the persistence of rare species. Ecology, 93, 456-461.