ANOSIM is for non-parametric data and operates on a ranked dissimilarity matrix, so is not suitable. But PERMANOVA or ANOVA is suitable :)

Thank you, that helps.

"The response variable is the proportional abundances of species in the estuary."

This sounds like as many response variables as you have species. Further, it may be affected by zero-inflation due to non-detection of rare species.

I would start with standardizing the data, ordination (e.g. Hellinger transformation based RDA) followed by cluster analysis, or reconsider response variables (e.g. once species richness, once evenness) and stick to ANOVA.

You have a very interesting real-world problem. If you are going to use parametric statistics be sure to test for equal variances as well as normality. Transformations may help normalize the data but often don't deal well with unequal variances. I also think you should consider Station to be a random variable because it represents all the locations within a section of the estuary. I can provide a reference for this if you want it but can't lay my hands on it right now.

I also suggest you look at the biology of the schooling species that are contributing to the high but sporadic catches. If they are a transient species such as down migrating juvenile Alosid species you could consider dropping them from your analyses because they probably are not good indicators of environmental impact. They are not around long enough to be good environmental indicators and just want to get out of the estuary. Often large catches of Atlantic Herring are removed from analyses of otter trawl data because they are likely captured near the surface during the haul back and are not a member of the demersal fish community. Removing species with high but sporadic catches may also help normalize your data. The best environmental indicators and integrators are species that can't move such as benthic inverts. If your schooling fish is a resident species than maybe you should retain them. Let the biology indicate how to handle your data.

Paul Geoghegan that is interesting and helpful. Between 75-92% of the total abundance of all my samples were mummichogs (Fundulus heteroclitus). Most species were residential or at least residents for the summer months. I would Love the reference on transformations if it is possible.

Rainer Ferdinand Wunderlich that is also helpful. I am hoping to analyze with a non-parametric multivariant test, as I want to identify what species are appearing to drive the differences... if there are any differences at all. I think Simper is the test i know about, but I am still trying to learning.

I will definitely take your advice for handling the non-parametric test.

My reference was for Station to be considered as a random variable. Both time and location (Station) of sampling can be considered random factors because both sampling date and selected locations represent only a fraction of all the possible times and locations. The reference is:

Underwood, A.J. 1994. On beyond BACI: Sampling designs that might reliably detect environmental disturbances. Ecological Applications. 4(1):3-15.

Mummichogs are a notoriously tolerant species and differences in abundance among stations might also be due to differences in physical habitat.

Sorry Andrew Paul McKenzie Pegman, you are quite wrong there! The fact that ANOSIM operates on ranks does not mean that it is not suitable. The whole approach in Primer is non-parametric but that means it is robust and very widely applicable. Whether you choose ANOSIM or Permanova for your hypothesis testing is partly a matter of preference. If you want to talk about significant interactions you need the latter, if you don't you may well get more meaningful results with the former.

To deal with the mummichog problem, try dispersion weighting (Clarke KR, Chapman MG, Somerfield PJ, Needham HR (2006) Dispersion-based weighting of species counts in assemblage analyses. Mar. Ecol. Progr. Ser. 320:11-27) prior to transforming the data - you may find that dispersion weighting on its own is sufficient, or that a weaker (e.g. square root) transformation will suffice. You may not need to convert to percentages if the net hauls are relatively standard.

ANOSIM can deal with ordered factors. The value of the statistic is a scaled measure of effect size (unlike Permanova). You don't need to worry about random or fixed factors.

As a good practice (recommended by PRIMER authors) I would suggest that you first do a visual analysis of your data by means of ordinations methos (nMDS, PCO) and see what the samples suggest in terms on (dis)similarity.

I hope this helps.

I was in a rush when I answered previously. I also wanted to say that you can't order 2 levels of a factor (only 3 or more) so I'd keep months as unordered. Locations down estuaries are probably ordered. I'd be inclined to go for a fully-crossed 3 way estuary (unordered), month (unordered), location (ordered) ANOSIM. For months I'd use a combined month-year factor, so you can include all observations. As Aldo says above, there are other exploratory options too, so make use of them.