Hi Pascal,

Never used myself but PERMANOVA seems possible from the PRIMER web site: see http://www.primer-e.com/permanova_details.htm

The procedure seems to be explained in details in the manual from page 64 on.

My apologies, I do not have time to check this more deeply this week as I'm returning home

Regards

Dom

Thanks Dom, I can't see how I missed this in the manual :-)

I will test it.

Pascal 

Hi Pascal,

The answer is yes. What I usually do is 1) a three factor PERMANOVA with Stie as a fixed factor, Time as a fixed or random factor depending on the hypothesis tested and Station (transect) as a random factor nested in site, or in site and time, depending on your hypothesis. Doing so you have a related sample design and the model "remove" the station effect form the error term which gives you amore powerful model. You can test each factor except station and the interaction time x site. If the PERMANOVA is significant I than do a CAP (Canonical Analysis of Principal Coordinate) with allows you to illustrate the structure using two group factors time and site. It a a constrained analysis which focus on the variation linked to the two group factors.

Permanova+ in Primer 7 can do all these analyses and much more...

Laurent

Thanks Laurent for your detailed answer... seems it is worth digging deeper !

I'll do a four factor PERMANOVA (adding the factor Year to Laurent's model) with Site as a random factor.

You can also explore the interactions with a second-stage MDS (see attached paper below).

Good luck!

Haizea

Thanks Haizea ; the 2nd-stage MDS is a good idea !

Pascal.

Hi All,

I'm just following this thread, as I have a similar analysis I would like to do.

I'm looking at patterns in fish assemblages over time, using a fully nested sampling design- reefs nested within management zones, sites nested within reefs, and fixed transects nested within sites. Surveys were done repeatedly over time in the fixed transects as per Pascals design, so I am also wondering how to approach PERMANOVA given its a repeated measures.

Laurent I am very interested in your approach in including transect in the PERMANOVA model. Just wondering how the model handles the fact that if transect is included as a factor, there would not be any residual component -i.e. no replication within the factor "transect". Or have I got my wires crossed with the specifics of the PERMANOVA model? Hope that makes sense :)

And can you elaborate on how this approach removes the effect of it being a repeated measures?

Many thanks!