I designed an experimental greenhouse study whereby I grew 67 plant
species from seeds collected from urban vacant land habitats in 2 soil
types (urban vacant lot and topsoil). Each species was replicated 10
times (10 pots each w/ an individual plant) in each soil type. I
measured a suite of continuous traits (i.e. height at maturity,
% germination success, aboveground biomass, belowground biomass and
specific leaf area) on each individual for each species over the duration
of the study. /
/
/
/My study question is based on understanding functional trait responses
of a pool of urban plant species to different soil environments. Because
trait responses can include both differences in mean values and/or
differences in the variances, and because these represent different
potential strategies for surviving in urban habitats, I would like to
test two types of models: the first is a model that tests the mean
differences between soil groups for each species (species is the unit of
replication) and the other testing differences in variance as response
variables. I am unsure, however, how to structure the model and I
cannot find any studies that have done anything comparable--not to say
they do not exist. /
/
/
/I am considering designing 2 generalized linear models that have soil
and functional group (a separate factor that I have not described in
this email) as explanatory variables and mean difference and difference
in variance/SD for each species as the response variables. My question is
as follows: /
/
/
/1) Is this possible and how would I structure each model? Specifically,
how do I represent mean differences and differences in variance,
separately? As I mentioned, I am interested in capturing general trends
in mean and variance across all species. Therefore, do I simply calculate and
compare effect size to capture mean differences (model 1) and F values
for a Levene?s test (or something comparable) (model 2)? Or, because sp.
is the unit of rep., would I have the mean and variance/SD for each sp. as
the response for each model and include sp. as a nested random effect?
For example-
lmer ( trait_mean ~ functional_group + soil + (1|species))
lmer ( trait_var ~ ?)