how to quantify the variable differences of GLM?

#1
Dear,

I am a student studying the relation between bird richness and environmental factors.

Here are three kinds of bird richness: A, B, C. (continuous variable)
and five sorts of environmental factors: 1, 2, 3, 4, 5. (continuous variable)

ordinary least square was used for model selection in General linear model.
The significant rank was as followed (based on delta AIC):

Model A 45312 (high to low)
Model B 45132
Model C 43512

Please advise
how to quantify the distinct impacts
each variable brings to each model ( for example, factor 3) ?

Is ANOVA suitable for comparison in this regard ?
(ANOCA considers interaction between variables but model selection does not).

Thank you very much.
 

bugman

Super Moderator
#2
You could try looking at canonical correspondence analysis fro your exploratory component and then fit a multiple regression model as your "formal analysis".

CANANCO and R do this nicely.
 
#5
Hello there,

I do not know of CANANCO, but there is a software called CANOCO that does CCA and other ordination/classification techniques... it might have been a typo, there ;)
However, I think CANOCO is not freeware, so you might have to pay for a licence. Besides that, it is quite frustrating to work with it in the beggining... it might need some getting used to.
If you are familiar with R, you could try package ade4, which has a 'cca' command with many options (and you can find tutorials online). However, it requires that you know a few things about the logic of R...

I hope this helps :)
 
#7
hello

thanks for the great help.
However, pls help answer what is the difference between CCA and PCA ?
I am familiar with PCA but need to learn CCA from scatch...
also, I found some guys using PCA in similar research

Thanks
 

bugman

Super Moderator
#8
PCA reduces the number of varaibles in a data set to a few "components" that best explain the bulk of the variation in the data. Assumes linearity so not good for species abundance data. PCA is a tricky one to explain, so if you need more, have a look online - there are several tutorial type documents.

Canonical Correspondence Analysis describes the relationship between species assemblages and environmental parameters - can include co-variates (sort of a combination of ordination and multiple regression). Assumes a unimodal gradient.

Canonical Correlation Analysis - uses two sets of data (e.g. water quality and environmental data) from the same object (e.g. site) - it assumes a linear gradient and DOES NOT differenciate between predictor and response varaibles.
 
#9
different impact of identical factors to distinct richness

Here are three kinds of bird richness: A, B, C. (continuous variable)
and five sorts of environmental factors: 1, 2, 3, 4, 5. (continuous variable)

You could try looking at canonical correspondence analysis fro your exploratory component and then fit a multiple regression model as your "formal analysis".

Here another question jumps out.

Is it possible to tell whether factor 1, 2, 3, 4, 5 bring different impact to richness A, B, C using canonical correspondence analysis and then multiple regression?