Data analyses: species size effects on — and responses to —
light penetration
Because there is no standard definition of “small species” we compared
two approaches: 1) small species were regarded as those smaller than the
first quartile of species height for all species in the focal community
(‘1st quartile’, <53.25 cm, N=15), 2) small species were
regarded as those smaller than the median of species height in the
community (‘median’, <83 cm, N=29; Fig. 2). Comparing results
across these two definitions allowed us to test whether any significant
patterns were sensitive to our method of defining small species.
We tested whether plots with larger species had lower light penetration.
Specifically, we fit linear models with mean plot height (mean
plot-level maximum height weighted by species plot-level abundance) and
large species abundance (including all species which are not considered
“small” by either definition, >83 cm) as predictors and
mean light penetration as the response variable. We also tested whether
small species abundance and richness responded to variation among plots
in light penetration. We fit linear models with small species abundance
and richness (for both ‘1st quartile’ and ‘median’
defined small species) as response variables. Light penetration and mean
intraplot variation in light penetration (calculated for each plot at
each sampling event and averaged across months) were our predictor
variables. We checked the variance inflation factor to ensure that there
was not a high degree of multicollinearity between our variables.
We used the package ‘stats’ from R.4.0.3 (R Core Team 2020) for
modeling, and checked statistical assumptions using residual vs. fitted,
normal quantile-quantile, scale location, and constant leverage plots
(‘ggfortify’ v.0.4.11; Tang 2016). We log10-transformed
variables where necessary to meet assumptions. For our multiple linear
regressions, we used the Akaike Information Criterion corrected for
small sample sizes for model selection (AICc). Where multiple models
were within two units of the lowest AICc score, we used the full model
average (via the ‘model.avg’ function from the ‘MuMIn’ v1.43.17 package;
Bartoń 2020) to determine the significance of predictors.
Finally, to assess the collective response of small species to light
availability, we performed distance-based redundancy analyses (dbRDA)
using Bray-Curtis dissimilarities. Additionally, we used the ‘decorana’
function from the package ‘vegan’ v.2.5-6 (Oksanen 2019) to confirm
linearity of responses. We conducted RDA with mean light penetration and
mean intraplot variation as possible predictors using the ‘capscale’
function from ‘vegan’ v.2.5-6 (Oksanen 2019). We compared models using
AICc scores, selecting the model with the lowest score.