Hi Veronica,

since you are working with both qualitative and quantitative traits an NMDS ordination would work to answer this question. Although more than asking "which traits are lost" I would ask "which trait values are favoured/hindered by disturbance" as the traits per se are very unlikely to dissapear. If working solely with quantitative traits, the best option would be to calculate average community-level trait values for each trait (abundance weighted, of course), and compare these values across your gradient . To be fully confident on your results I would perform both analyses (NMDS with both quantitative/qualitative traits and community-averaged trait values) and see if they lead to the same conclusions

Some recent works with relatively similar approaches that you might be interested in are:

Frenette-Dussault et al. 2012 JEcol

Gross et al. 2013 JEcol

Sandra Lavorel or David Ackerly are authors that definetly you want to look at

hope it helps

Hi Veronica,

since you are working with both qualitative and quantitative traits an NMDS ordination would work to answer this question. Although more than asking "which traits are lost" I would ask "which trait values are favoured/hindered by disturbance" as the traits per se are very unlikely to dissapear. If working solely with quantitative traits, the best option would be to calculate average community-level trait values for each trait (abundance weighted, of course), and compare these values across your gradient . To be fully confident on your results I would perform both analyses (NMDS with both quantitative/qualitative traits and community-averaged trait values) and see if they lead to the same conclusions

Some recent works with relatively similar approaches that you might be interested in are:

Frenette-Dussault et al. 2012 JEcol

Gross et al. 2013 JEcol

Sandra Lavorel or David Ackerly are authors that definetly you want to look at

hope it helps

The ordination will help you identify the traits that are being favoured by your species. You would also need to have the type/characteristics of the responding species so that you will be able tell the direction of change. And I agree with Santiago Soliveres to also use community level trait values

Thank you both for your very helpful answers!! Now I´m thinking about assessing how traits change along the gradient by calculating CWM of each trait (instead of which are lost). I could transform the qualitative traits to ordinal variables in order to analyze all traits, not only the quantitaive onces.

Hi Veronica,

Sounds interesting! Check out this new trait based approach for examining community responses to disturbance recently proposed

http://www.sciencedirect.com/science/article/pii/S0169534712002650

I just submitted a paper using these methods to quantify fish community responses to climate extremes. Super easy to run, very informative, and produces nice visually appealing figures!

Here is another paper that may help you choose an analysis!

http://onlinelibrary.wiley.com/doi/10.1111/j.1461-0248.2010.01444.x/full

There are all the work of J. Bremner that can help you. For example :

Bremner J., 2005. Assessing ecological functioning in marine benthic communities. Newcastle University, 220p.

Bremner J., Rogers S. I. & Frid C. L. J., 2006. Methods for describing ecological functioning of marine benthic assemblages using biological traits analysis (BTA). Ecological Indicators. 6, 609-622.

There is also this article which approximates your problem:

Wooster DE, Miller SW & DeBano SJ, 2011. An Examination of the Impact of Multiple Disturbances on a River System: taxonomic Metrics Versus Biological Traits. River Research and Applications.

There is possibility of coupled matrices "biological traits by species" and "abundance of species per site (or along a gradient)"

An interesting approach is the statistical analysis RQL that connects three data matrices. This analysis allows the coupling matrices "biological traits by species," "abundance of species per site" and "environmental characteristics by sites"

Hi Veronica,

we explored the multivariate relationship between plant species traits, stage of forest succession after disturbance and environmental gradients. We used a recently developed data-analytical method, Model-Based Recursive Partitioning, to disentangle the relationships between patterns of plant traits and environmental gradients. And I guess that such approach can help also in your specific case!

The paper is: Campetella et al. 2012 - Patterns of plant trait–environment relationships along a forest succession chronosequence. Agriculture Ecosystems and Environment, 145, 38-38. you can download the paper from my personal RG page.

cheers

Giandiego Campetella

Hi, Veronica,

I have written some seriation code based on the assumption that the gradient response curve is unimodal. If if you suspect that this could be the case of your data, we can give it a try.

Kindest regards,

Peter.

Hi, Veronica,

I forgot the second answer:

Due to violation of assumption of linearity, many ordination methods are incorrect from the statistical point of view. Whether they +/- work is another question.

Best,

Peter.

Hi Peter, but my data does not have an unimodal distribution, mainly because the grazing gradient was measured in a piosphere (radial degradation from a watering point).

Hi Veronica,

by unimodal distribution of traits along a gradient I meant that some may have a peak at low values, others in medium grazing, and some at high levels...

Best,

Peter.

Hi Veronica,

I agree with Santiago, using ordination is appropriate but really tries to answer the question what is the relationship of the traits to the disturbance gradient and as long as you have a strong gradient, there should be traits that are low at both ends and high at both ends. Thus these traits should be the ones that you evaluate as being lost or increased due to the disturbance. Another way to check this directly is to break your sites up into "good or unimpaired" and "disturbed" or Low, Medium, High, then using the similarity matrix based on the trait information perform an ANOSIM test of difference between these groups (you can make 3-4 or as many groups as you want), this will not only tell you whether the groups are statistically different but also what traits explain these differences among groups. This might give you more of a direct assessment of your original question, "which traits are lost". Btw, if only two groups, then a Chi-square test would be the best statistical test, but it won't tell you which traits are related to those differences.

Hope this helps,

Ian