basic stuff Flashcards
Direct, long gradient
CCA
Direct, short gradient
RDA
Constrained
CCA and RDA
unconstrained
CA and PCA, distance based approaches (mds and polar)
Indirect, long gradient
CA
Indirect, short gradient
PCA
Parametric (linear) distance based approach
Classical MDS
Non-parametric (ranks) distance based approach
nonmetric MDS
when do you add environmental data afterwards?
indirect ordination
long gradient response?
unimodal
short gradient response?
linear
What do Log (1+x) and sqrt(x) transformations do?
Reduce the influence of large values
What does the Hellinger transformation do?
computing sqrt of relative abundances per location
What does the Wisconsin transformation do?
‘double standardization’: first computing relative
abundances per location, then normalizing values to max
value=1 per species
which distances are symmetric?
Manhattan and Euclidian
which distances are asymmetric?
chi-squared and Bray-Curtis
what is inertia for CCA/CA
Chi-square
what is inertia for RDA/PCA
Variance
what distance is CA based on?
Chi-squared distance
what is the average MDS by design?
0
when is stress level acceptable?
<0.05 is excellent, <0.1 is great, <0.2 is good/ok, <0.3 is poor
how are distances mapped in Non parametric MDS?
The ranks are mapped
is non paramteric MDS the same result every time you try?
No, it’s iterative (each run may have different results)
how to check the quality of NMDS?
Inspecting stress value, Shepard diagram
how to choose how many dimensions for NDMS?
The least amount of dimensions where the stress is still below 0.1 (more dimensions reduce stress but you want the least amount of dimensions)
how to choose how many axis for PCA?
scree plot with broken stick method
what is the total inertia for CA?
Average chi-squared distance among all sites, or sum of all eigenvalues
what do eigenvalues represent for CA?
the chi-square value per axis
what distance is PCA based on?
Linear species response based on Euclidian distance
when is a gradient long or short?
first axis of a DCA (detrended) is 3 or <3 it is short. Also if there are many zeroes it is most likely long, and most species in most sites means short.
how to know in output if a test was constrained (CCA/RDA) or not (CA/PCA)?
if output does not say ‘constrained’ it is unconstrained
how to calculate proportion explained by an axis?
eigenvalue of axis / total inertia
in a significance test for constrained ordination, what refers to restrained and unrestrained part?
residual= unrestrained. Model = restrained
when is inertia chi-square?
for CA/CCA
when is inertia variance?
PCA/RDA
what is total inertia for PCA?
sum of the eigenvalues of all axes, or sum of all variance (since inertia = variance)
what does it mean when the outcome of an unconstrained and constrained analysis are very different?
the enviornmental variables do not explain the variance well
when can you use multivariate analysis
for anything that has multi-response data that allowes calculating dissimilarities
in an constrained output how many axes are used?
the constrained + unconstrained ranks. constrained = amount of environmental variables
when do you have to check for colinearity of explanatory variables?
with CCA and RDA because they are regression methods
when is collinearity a problem?
vif >5 potentially, >10 definitetly
what is conditioning?
removing the effect of gradients in a landscape for further testing
what are canonical axes?
constrained axes
what is a marginal effect?
the effect of a particular term when all other model terms are included in the model
how are marginal terms tested?
by eliminating each term from the model containing all other terms
are marginated effects dependent on order of the terms?
no, but correlated terms will get high P values
does the order matter when testing terms?
yes, if there is correlation
what is the P value in a permutation test?
the fraction of permutation values that are larger than the actual value
how is p value calculated in permutation test?
the number of values that are higher than the actual value, divided by (the amount of permutations + 1)
in an unconstrained ordination do the eigenvalues increase or decrease over the axes?
decrease
when do you use chi-square distance
when long gradient and rare species are well sampled
when do you use Euclidian or mannhattan distance
when there’s a short gradient
when do you use bray-curtis disctance
long gradient and rare species not well sampled
what to be careful of with Manhattan/euclidian distance?
Scaling, double zeroes, abundance paradox
what to be careful of with Bray curtis distance?
Large values
what to be careful of with Chi square distance?
rare species (that aren’t sampled well), distance between two sites is influenced by other sites
