Soil characteristics were analyzed 4 yr after experiments were
implemented, so data should be interpreted cautiously because nutrient
values may have changed. We took samples at 0–10 cm depth
for five gaps and five understory sites of eight studied in each fragment
(Table S1). Overall, the burned fragment (L3) had the highest
levels of soil nutrients and L2 had the highest proportion of sand.
Gaps had higher concentrations of Ca and Na than understory sites.
The gaps were created by selective logging (or naturally in U)
1–3.5 yr prior to the experiment. We managed gaps before planting
by cutting saplings >2 m in height (residual saplings from pregap
formation) to standardize gap structure within and among forest
fragments. Smaller vegetation was disturbed considerably during
the experiment establishment (cutting saplings, establishing fences,
digging and planting seedlings). To prevent cattle access to half of
the seedlings, we built sturdy 3 × 6 m enclosures from wood posts
and wire in each gap and understory site. One seedling of each
species was planted inside and one outside of these fenced areas.
Seedlings were planted on a 1 × 1 m grid positioned in the center
of the gap. Thus, the experimental design was: 7 species × 2 canopy
conditions × 8 replicates × 2 levels of cattle access × 4 fragments
× 1 seedling = 896 seedlings (128 seedlings per species). Seedlings
were measured 1 wk after planting for height of apical meristem and
diameter at stem base, and then monthly for 1 yr. When possible,
the cause of mortality was determined.
Survival after 1 yr was analyzed as a function of the main effects
(species, habitat, and cattle enclosure) using log-linear analysis
(Statsoft 2000, Tabachnick & Fidell 2001). Data from each fragment
were analyzed independently because logging effects were not
replicated. We verified the partial association of each factor and
interaction, computing the likelihood ratio χ2 of the model that
included all main factors with the model that excluded each main
factor, and then repeated the process for all two-way interactions
(Statsoft 2000). To select the best model we started with a model
including all significant partial associations and then eliminated factors
and interactions that did not improve the model at P < 0.05
(Statsoft 2000).
To test the hypothesis that mortality in gaps is higher during
the dry season, we used survival analysis (Statsoft 2000). We compared
the survival curves, which describe the proportion of survivors
during a period as a fraction of the number alive at the beginning
of the period. The Wilcoxon test was used to compare observed to
expected numbers of failure in each interval (Statsoft 2000).
Growth data were analyzed as Relative Growth Rate (RGR),
calculated as: RGR (%) = (lnH(t2) – lnH(t1) / t2 – t1) × 100; where
H = height, t = time in years.
For each fragment, we ran a split-plot ANOVA to test the
effects of habitat (whole plot), species (split plot), and their interactions
on growth in height and diameter using the PROC
MIXED procedure in SAS (Littell et al. 1996). In cases of significant