Ecological changes in echinoderm populations: what statistical model is appropriate

#1
Hi Talk Stats and associated folks

We are doing an ecological study on the effects of removal of an invasive seastar (Asterias amurensis) from Port Phillip Bay Vic, Aus. We are monitoring 2 sites across 3 habitat types, using 2 belt transects in each habitat site. We are monitoring 4 species Three native seastars and Asterias at both sites. what we want to analyse is the change in species over time to determine if the removal of the pest effects native species abundance. We are thinking we need to do T-tests and then perform a nested anova or multivarite anova on significant results.

anyone who would like to help us out, your time will be much appreciated

THank you

Jason
 
#2
Re: Ecological changes in echinoderm populations: what statistical model is appropria

Hi Jason.

If habitats within sites and belts within habitats can be assumed indepent, my first thought would be to conduct a glm with Poisson errors. But I'm not sure, you have given too little information.

What is the protocol for including seastar's; only when they are on the belt - or is the belt the central location for distance sampling?

Anyway a Poisson glm is more appropriate for count data, but I am assuming you are conducting repeated sampling over time? How many repeat visits are there?

If you provide more information, we will be in a better position to help. Maybe even our resident aquatic ecologist, bugman, will respond?

Cheers,

TE
 

bugman

Super Moderator
#3
Re: Ecological changes in echinoderm populations: what statistical model is appropria

I agree with TE's approach. But are you looking at presence/absence (binomial) or count data?

Generalised estimating equations might help you with the temporal aspects if there is autocorrelation. If you are interested in variance partioning of the various components of the study, you might consider nested models. But if you are interested in simialrities in species assembalges over time there are also many option available using multivariate techiniques...
 
Last edited:
#4
Re: Ecological changes in echinoderm populations: what statistical model is appropria

Thanks

Habitats are the same each time we sample and the transect lines are placed in the same locations each time we have sampled. we have repeated the samples 4 times in each location. seastar counts are completed along the transect with 1 m either side of the transect included. our main focus is to determine if there is a change in native species counts after asterias amurensis has been removed

kind regards

jase
 
#5
Re: Ecological changes in echinoderm populations: what statistical model is appropria

Hi Jase,

I don't know what bugman would suggest but to me it seems that the repeat visits won't be independent. So you have two options (if belts within habitats are independent that is!);

1) use a mixed effect model. Try a GLMM (generalized linear mixed model) with Poisson errors and fixed effect can be site, with the repeated visits as random effects.
2) Calculate mean density over visits, and conduct an Anova analysis. Your sample size here is 12 (2*3*2 = 12, so 6 per site.. which isn't much).

Does this help?

TE
 

bugman

Super Moderator
#6
Re: Ecological changes in echinoderm populations: what statistical model is appropria

Unfortuantly, I agree with TE here. I say unfortunalty because GLMM's can be tricky to grasp so I would strongly recommend some background reading, and this:


http://www.ncbi.nlm.nih.gov/pubmed/19185386 (if I find my pdf I'll post it - but cant just now)

is a great paper to start with (one of the authors is from my old department - no bias implied).

Habitat type will also be a fixed effect!

Another approach is to look at a PERMANOVA type analysis. In your case something like Jaccards simialrity measurement would give you in percent differences in species assemblys over time. Just a thought, but this might not directly address your hypothesis.
 
#7
Re: Ecological changes in echinoderm populations: what statistical model is appropria

Hi TE and Bugman

Our belt transects are placed 3 meters apart from each other so there would not be crossover of data collection between the transects, within habitats, therefore can we assume that they are in fact independent of each other? And continue with the GLMM. Due to our small sample size would this method be preferred to an anova model and be more descriptive of the data?

Our null Hypothesises are:
1) The removal of Asterias from sites has no siginificant effect on future abundance within that site
2) The removal of Asterias amurensis has no significant effect on the abundance of Native seastar species

Would these suit a PERMANOVA design Bugman? Or would they be better described through the GLMM?
I believe that both methods can be completed in SPSS which I have available to use for analysis processing

Thankyou very much for your assistance it is amazing some of this stats can get complicated :)

Coastalresearch
 
#8
Re: Ecological changes in echinoderm populations: what statistical model is appropria

also we are not comparing the sites to each other just looking at the changes within 2 seperate sites as they are significantly distanced apart we can assume that they would have different assemblages anyhow, in terms of doing a GLMM will my data set have to be set up for individual results per species or can the GLMM pull apart that data for example can I have my data set up:


Sample Habitat Species 1 Species 2 Species 3
1 1 0 6 19
1 2 3 4 8
1 3 6 10 27

or does the data need to be separated per species

Sample Habitat A Habitat R Habitat SS Abundance A Abundance R Abundance SS
1 1 2 3 0 78 19
2 1 2 3 2 27 0
3 1 2 3 4 103 2
 
Last edited: