Carrying out MDS and Clarke and Warwick's BIOENV analysis using Primer version 5

Version 5 of Primer, which is a Windows based programme is on the first 10 computers in Lab D. It is much more user friendly than version 4, and can import a wider range of data files. In particular, it can read Excel spreadsheets, including importing species and sample names. For example, to import the following data in an Excel file:

  sp1 sp2 sp3
site1 1 6 3
site2 4 3 6
site3 2 1 9

Select File/Open, move to the directory that contains your data file and set the file type as Excel files. You should see the file displayed. The default option is to read data on samples, so accept this. For this format of file, make sure that the "Includes row labels" box is ticked, and that the "Includes title" box is not ticked. Click on OK, then on the next dialogue box enter a title for your data, choose the Samples as Rows radio button and check that the number of columns and rows is correct. It is then worth using the Save As command to save the data in a Primer format data file.

The file is available here if you want to try. If your data are not already in an Excel file, you can create a new data sheet within Primer, and type in data directly. You can also put samples into groups by defining one or more factors (Select the pull down Edit menu then choose Factors) - there are needed if you want to use the ANOSIM procedures, and can also be used to distinguish data points on graphs. Unlike SPSS, Primer places data on environmental variables into a separate spreadsheet from the data on species abundances.

MDS

It is possible to do a PCA in Primer, but the procedure is less flexible than in SPSS. The main things that we will use Primer for in this course are to carry out a multidimensional scaling (MDS) and to examine relationships between environmental variables and ecology. To carry out an MDS, first select Similarity from the Data menu, and choose which transformation and similarity measure you want to use, and whether or not you want to standardise your data (give equal weight to each species). We will usually want to work on unstandardised data as this gives a greater weighting to the species with the highest variance (comparable to doing PCA on the covariance matrix). Euclidean distance will give comparable results to PCA and K-means clustering, although in the marine literature a Bray-Curtis similarity is often used as it assesses similarity between two samples on the basis of species that occur in both and species that occur in on only but does not take the absence of a species in both samples as a sign of similarity. Then with the resulting similarity matrix in the active window, select Analyse, then MDS. Set the number of restarts (don't choose too high a value, as it has to think hard if you have a large data set). Primer will produce a 2D plot, with samples indicated by their labels. But with the graph in the active window, you can select Graph then Properties and change this to a 3D plot, change symbols or labels plotted or produce a 2D "Bubble Plot" in which the size of the symbols is proportional to the magnitude of an environmental variable

Permutation tests

Primer allows you to do a number of permutation tests - in effect non-parametric tests of hypotheses about multivariate data. These include:

Anosim - Comparable to one- and two-way analysis of variance. This allows you to test questions like: ·

To carry out these analyses, you need to define Factors (see above) nefore running the ANOSIM procedure.

BioEnv and BVStep. These carry out tests of the relationship between community composition and environmental variables - comparable to multiple regression. BIOENV examines all possible combinations of variables, from each environmental variable separately through to all at the same time. If you have large numbers of environmental variables, this can be very time consuming so BVSTEP first fits the environmental variable with the strongest relationship, then adds in the variable with the next strongest relationship etc, comparable to stepwise regression.

To examine relationships between ecological and environmental data using the BIOENV procedure, have the table of environmental variables in the active window, then choose BIOENV from the Analyse menu:

Primer does a whole lot of other things, including producing a number of nice graphical displays and looking at the contribution that each species makes to the differences between groups. But if you want to do more that the basic analyses referred to here, please consult the manual, the accompanying book "Change in Marine Communities" or the online help. The package is very easy to use once you understand the statistical methods.

Alastair Grant

29th November 2002