3m ago
8.95 MB
262 Pages
Last View : 1m ago
Last Download : 1m ago
Upload by : Giovanna Wyche

CHANGE IN MARINE COMMUNITIES:An Approach to Statistical Analysisand Interpretation3rd editionK R Clarke, R N GorleyP J Somerfield & R M Warwick

CHANGE IN MARINE COMMUNITIES:An Approach to Statistical Analysis andInterpretation3rd editionK R Clarkea,b, R N Gorleya, P J Somerfieldb& R M Warwickba(Formerly) PRIMER-E Ltd, PlymouthbPlymouth Marine Laboratory

Published 2014byPRIMER-E LtdBusiness Address: 3 Meadow View, Lutton, Ivybridge, Devon PL21 9RH, United KingdomReproduced from 15 August 2016byPRIMER-e (Quest Research Limited)Business Address: ecentre, Gate 5 Oaklands Rd, Massey University Albany Campus, Auckland, New ZealandFirst edition 1994Second edition 2001Third edition 2014Citation:Clarke, K.R., Gorley, R.N., Somerfield, P.J., Warwick, R.M. (2014)Change in marine communities: an approach to statistical analysis and interpretation,3nd edition. PRIMER-E: Plymouth Copyright 2014, all rights reserved

CONTENTSINTRODUCTION and ACKNOWLEDGEMENTSCHAPTER 1A framework for studying changes in community structureCHAPTER 2Simple measures of similarity of species ‘abundance’ between samplesCHAPTER 3Clustering methodsCHAPTER 4Ordination of samples by Principal Components Analysis (PCA)CHAPTER 5Ordination of samples by Multi-Dimensional Scaling (MDS)CHAPTER 6Testing for differences between groups of samplesCHAPTER 7Species analysesCHAPTER 8Diversity measures, dominance curves and other graphical analysesCHAPTER 9Transformations and dispersion weightingCHAPTER 10 Species aggregation to higher taxaCHAPTER 11 Linking community analyses to environmental variablesCHAPTER 12 Causality: community experiments in the field and laboratoryCHAPTER 13 Data requirements for biological effects studies: which components andattributes of the marine biota to examine?CHAPTER 14 Relative sensitivities and merits of univariate, graphical/distributional andmultivariate techniquesCHAPTER 15 Multivariate measures of community stress, and relating to modelsCHAPTER 16 Further multivariate comparisons and resemblance measuresCHAPTER 17 Biodiversity and dissimilarity measures based on relatedness of speciesCHAPTER 18 Bootstrapped averages for region estimates in multivariate means plotsAPPENDIX 1Index of example dataAPPENDIX 2Principal literature sources and further readingAPPENDIX 3References cited

Introductionpage 0–1INTRODUCTIONThird editionAttribution (and responsibility for queries)The third edition of this unified framework for nonparametric analysis of multivariate data, underlyingthe PRIMER software package, has the same formand similar chapter headings to its predecessor (withan additional chapter). However, the text has beenmuch expanded to include full cover of methods thatwere implemented in PRIMER v6 but only describedin the PRIMER v6 User Manual, and also the entirerange of new methods contained in PRIMER v7.These new sections have all been authored by KRCbut build heavily on collaborations, joint publicationsand novel algorithmic and computer coding workwith/by PJS and RNG. In the retained material fromthe 2nd edition (authored by KRC and RMW), KRCwas largely responsible for Chapters 1-7, 9, 11 and 16and RMW for 10 and 12-14, with the responsibilityfor Chapters 8, 15 and 17 shared between them.Whilst text has been altered throughout, PRIMER v6users familiar with the 2nd edition, who just want tolocate the new material, will find it below:Table 0.1. Manual pages primarily covering new materialTopicsPagesAdditions to the framework1-9 to 13Missing data and variable weightings2-9 to 10Similarity profile tests (SIMPROF) ofclusters on sample dendrogramsUnconstrained binary divisive (UNCTREE)and fixed group (k-R) clusteringMore nMDS diagnostics (MST, similarityjoins, 3-d cluster on MDS, scree plots)Metric MDS (mMDS), threshold MDSCombined MDS (‘fix collapse’ by nMDS mMDS, composite biotic/abiotic nMDS)ANOSIM for ordered factors3-way ANOSIM designsSpecies Analyses (new chapter, in effect):SIMPROF on species (coherent curves)Shade plots ( dendrograms, axes orders)Bubble plots (for groups, segmented)Testing curves (dominance/particle/growth)Analysing multiple diversity indicesDispersion weightingVector plots in PCA and MDSGlobal BEST test (allowing for selection)and constrained BEST analysesLinkage trees: binary clusters, constrainedby abiotic ‘explanations’ (LINKTREE)Model matrices, RELATE tests of seriationand cyclicity, constrained RELATESecond-stage analysis (2STAGE)Zero-adjusted Bray-Curtis for sparse dataDefining and comparing resemblancesSecond-stage ‘interaction’ plotsTaxonomic (relatedness-based) dissimilarityMeans plots & ‘bootstrap average’ regions3-6 to 103-10 to 145-7, 5-11,5-135-14 to 175-18 to 206-14 to 176-18 to 267-1 to 77-7 to 147-17 to 208-13 to 158-15 to 169-5 to 1011-2, 11-511-10 to 1211-13 to 1615-10 to 1416-7 to 916-10 to 1316-14 to 1817-18 to 2018-1 to 8PurposeThis manual accompanies the computer softwarepackage PRIMER (Plymouth Routines In MultivariateEcological Research), obtainable from PRIMER-e,(see Its scope is the analysis ofdata arising in community ecology and environmentalscience which is multivariate in character (manyspecies, multiple environmental variables), and it isintended for use by ecologists with no more than aminimal background in statistics. As such, this methodsmanual complements the PRIMER user manual, bygiving the background to the statistical techniquesemployed by the analysis programs (Table 0.2), at alevel of detail which should allow the scientist tounderstand the output from the programs, be able todescribe the results in a non-technical way to othersand have confidence that the right methods are beingused for the right problem.This may seem a tall order, in an area of statistics(primarily multivariate analysis) which has a reputationas esoteric and mathematically complex! However,whilst it is true that the computational details of someof the core techniques described here (for example,non-metric multidimensional scaling) are decidedly nontrivial, we maintain that all of the methods that havebeen adopted or developed within PRIMER are soconceptually straightforward as to be amenable tosimple explanation and transparent interpretation. Infact, the adoption of non-parametric and permutationapproaches for display and testing of multivariate datarequires, paradoxically, a lower level of statisticalsophistication on the part of the user than does a satisfactory exposition of classic (parametric) hypothesistesting in the univariate case.One primary aim of this manual is therefore to describea coherent strategy for the interpretation of data oncommunity structure, namely values of abundance,biomass, % cover, presence/absence etc. for a set of‘species’ variables and one or more replicate samples

Introductionpage 0–2Table 0.2. Chapters in this manual in which the methods underlying specific PRIMER routines are principally e for samplesAssociation index for speciesDummy variables (zero-adjusted coefficient)Taxonomic dissimilaritiesClusterCLUSTER (hierarchical: agglomerative)LINKTREE (“:constrained divisive)UNCTREE (“ :unconstrained divisive)kRCLUSTER (non-hierarchical)Clustering variables (species)SIMPROFtests for sample groups from Clustertests for species groupsCoherence plots (Line plots)PCA ( Vector plot)MDSNon-metric, Metric, Combined, Fix collapseShepard diagram, Scree plotOverlay clusters, trajectory, MST, join pairsVector plotBubble plots (groups, multiple)ANOSIM (1/2/3-way, crossed/nested, ordered)SIMPERShade Plot (Matrix display)Diversity indicesDIVERSECASWELL, Geometric Class PlotsDominance Plots, DOMDISSpecies Accumulation PlotsTAXDTEST, histogram/funnel/ellipse plotsPre-treatmentTransform, StandardiseNormalise VariablesCumulate SamplesDispersion Weighting, Variability WeightingAggregateBESTBIO-ENV, Draftsman PlotConstrained BEST (Within factor levels)BVSTEP, Global BEST testMVDISPRELATE (Seriation, Cyclicity, Model Matrix)2STAGE (Single and Multiple matrices)Bootstrap 618PRIMER has a range of other data manipulation and plottingroutines: Select, Edit, Summary stats, Average, Sum, Transpose,Rank, Merge, Missing data and Bar/Box/Means/Scatter/Surface/Histogram Plots, etc – see the PRIMER User Manual/Tutorial.which are taken:a) at a number of sites at one time (spatial analysis);b) at the same site at a number of times (temporalanalysis);c) for a community subject to different uncontrolledor controlled manipulative ‘treatments’;or some combination of these.These species-by-samples arrays are typically quitelarge, and usually involve many variables (p species,say) so that the total number (n) of observed samplescan be considered to be n points in high-dimensional(p-dimensional) space. Classical statistical methods,based on multivariate normality are often impossibleto reconcile with abundance values which are predominantly zero for many species in most samples,making their distributions highly right-skewed. Evenworse, classic methods require that n is much largerthan p in order to have any hope of estimating theparameters (unknown constants, such as means andvariances for each species, and correlations betweenspecies) on which such parametric models are based.Statistical testing therefore requires methods whichcan represent high-dimensional relationships amongsamples through similarity measures between them,and test hypotheses without such model assumptions(non-parametrically within PRIMER by permutation).A key feature is that testing must be carried out onthe similarities, which represent the true relationshipsamong samples (in the high-d space), rather than onsome lower-dimensional approximation to this high-dspace, such as a 2- or 3-d ‘ordination’.Data visualisation, however, makes good use of suchlow-dimensional ordinations to view the approximatebiological relationships among samples, in the formof a ‘map’ in 2- or 3-d. Patterns of distance betweensample points in that map should then reflect, asclosely as possible, the patterns of biological dissimilarity among samples. Testing and visualisationare therefore used in conjunction to identify andcharacterise changes in community structure in timeor space, and in relation to changing environmental orexperimental conditions.Scope of techniquesIt should be made clear at the outset that the title‘Change in Marine Communities’ does not in anyway reflect a restriction in the scope of the techniquesin the PRIMER package to the marine environment.The first edition of this manual was intended primarilyfor a marine audience and, given that the examplesand rationale are still largely set around the literatureof marine ecology, and some of the original chapters

Introductionpage 0–3in this context have been retained, it seems sensibleto retain the historic continuity of title. However, itwill soon be evident to the reader that there is ratherlittle in the methods of the following pages that isexclusively marine or even confined to ecology. Infact, the PRIMER package is now not only used inover 120 countries world-wide (and in all US states)for a wide range of marine community surveys andexperiments, of benthic fauna, algae, fish, plankton,corals, dietary data etc, but is also commonly foundin freshwater & terrestrial ecology, palaeontology,agriculture, vegetation & soil science, forestry, bioinformatics and genetics, microbiology, physical(remote sensing, sedimentary, hydrological) andchemical/biochemical studies, geology, biogeographyand even in epidemiology, medicine, environmentaleconomics, social sciences (questionnaire returns), onecosystem box model outputs, archaeology, and so on¶.Indeed, it is relevant to any context in which multiplemeasurement variables are recorded from each sampleunit (the definition of multivariate data) and classicalmultivariate statistics is unavailable, i.e. especially (asintimated above) where there are a large number ofvariables in relation to the number of samples (and inmicrobial/genetic studies there can be many thousandsof bands with intensities measured, from each sample),or characterised by a presence/absence structure inwhich the information is contained at least partly inpattern of the presences of non-zero readings, as wellas their actual values (in other words, data for whichzero is a ‘special’ number).As a result of the authors’ own research interests andthe widespread use of community data in pollutionmonitoring, a major thrust of the manual is the biologicaleffects of contaminants but, again, most of the methodsare much more generally applicable. This is reflectedin a range of more fundamental ecological studiesamong the real data sets exemplified here.¶The list seems endless: the most recent attempt to look at whichpapers have cited at least one of the PRIMER manuals, or a highlycited paper (Clarke 1993) which lays out the philosophy and somecore methods in the PRIMER approach, was in August 2012, andresulted in 8370 citations in refereed journals (SCI-listed), from773(!) different journal titles. Of course, there is no guarantee thata paper citing the PRIMER manuals has used PRIMER – thoughmost will have – but, equally, there are several score of PRIMERmethods papers that may have been cited in place of the manuals,especially for the many PRIMER developments that have takenplace since the Clarke (1993) paper, so the above citation total islikely to be a significant underestimate. A listing of these journals,and their frequency of PRIMER-citing papers, together with thereference list of the 8370 citing papers, can be downloaded fromthe PRIMER-E web site, This list can have afunction in searching for past PRIMER-citing papers in the user’sown discipline, a support question often asked of PRIMER-E staff.The literature contains a large array of sophisticatedstatistical techniques for handling species-by-samplesmatrices, ranging from their reduction to simple diversity indices, through curvilinear or distributionalrepresentations of richness, dominance, evenness etc.,to a plethora of multivariate approaches involvingclustering or ordination methods. This manual doesnot attempt to give an overview of all the options.Instead it presents a strategy which has evolved overdecades within the Community Ecology/Biodiversitygroups at Plymouth Marine Laboratory (PML), andsubsequently within the ‘spin-out’ PRIMER-E Ltdcompany, and which has now been tested for ease ofunderstanding and relevance to analysis requirementsat well over 100 practical 1-week training workshops.The workshop content has continued to evolve, inline with development of the software, and the utilityof the methods in interpreting a range of communitydata can be seen from the references listed underClarke, Warwick, Somerfield or Gorley in Appendix 3,which between them have amassed a total of 20,000citations in SCI journals. The analyses and displays inthese papers, and certainly in this manual, have verylargely been accomplished with routines available inPRIMER (though in many cases annotations etc havebeen edited by simply copying and pasting into graphicspresentation software such as Microsoft Powerpoint).Note also that, whilst other software packages willnot encompass this specific combination of routines,several of the individual techniques (though by nomeans all) can be found elsewhere. For example, thecore clustering and ordination methods describedhere are available in many mainstream statisticalpackages, and there are at least two other specialisedstatistical programs (CANOCO and PC-ORD) whichtackle essentially similar problems, though usuallyemploying different techniques and strategies; otherauthors have produced freely-downloadable routinesin the R statistical framework, covering some of thesemethods.This manual does not cover the PERMANOVA routines, which are available as an add-on to thePRIMER package. The PERMANOVA software hasbeen further developed and fully coded by PRIMERE (in the Microsoft Windows ‘.Net’ framework of allrecent PRIMER versions) in very close collaborationwith their instigator, Prof Marti Anderson (MasseyUniversity, NZ). These methods complement those inPRIMER, utilising the same graphical/data-handlingenvironment, moving the emphasis away from nonparametric to semi-parametric (but still permutationbased and thus distribution-free) techniques, whichare able to extend hypothesis testing for data with

Introductionpage 0–4more complex, higher-way designs (allowing, forexample, for concepts of fixed vs random effects, andfactor partitioning into main effect and interactionterms). This, and several other analyses which moreclosely parallel those available in classical univariatecontexts, but are handled by permutation testing, arefully described in the combined Methods and Usermanual for PERMANOVA , Anderson (2008).Example data setsThroughout the manual, extensive use is made of datasets from the published literature to illustrate the techniques. Appendix 1 gives the original literature sourcefor each of these 40 data sets and an index to all thepages on which they are analysed. Each data set isallocated a single letter designation (upper or lowercase) and, to avoid confusion, referred to in the textof the manual by that letter, placed in curly brackets(e.g. {A} Amoco-Cadiz oil spill, macrofauna; {B} Bristol Channel, zooplankton; {C} Celtic Sea,zooplankton, {c} Creran Loch, macrobenthos etc).Many of these data sets (though not all) are madeavailable automatically with the PRIMER software.Literature citationAppendix 2 lists some background papers appropriateto each chapter, including the source of analyses andfigures, and a full listing of references cited is givenin Appendix 3. Since this manual is effectively a book,not accessible within the refereed literature, referral tothe methods it describes should probably be by citingthe primary papers for these methods (this will notalways be possible, however, since some of the newroutines in PRIMER v7 are being described here forthe first time). Summaries of the early core methodsin PRIMER for multivariate and univariate/graphicalanalyses are given respectively in Clarke (1993) andWarwick (1993). Some primary techniques papersare: Field et al (1982), for clustering, MDS; Warwick(1986) and Clarke (1990), ABC and dominance plots;Clarke and Green (1988), 1-way ANOSIM, transformation; Warwick (1988b) and Olsgard et al (1997),aggregation; Clarke and Ainsworth (1993), BEST/Bio-Env; Clarke (1993) and Clarke and Warwick(1994), 2-way ANOSIM with and without replicates,similarity percentages; Clarke et al (1993), seriation;Warwick and Clarke (1993b), multivariate dispersion;Clarke and Warwick (1998a), structural redundancy,BEST/BVStep; Somerfield and Clarke (1995) andClarke et al (2006b), second-stage analyses; Warwickand Clarke (1995, 1998, 2001), Clarke and Warwick(1998b, 2001), taxonomic distinctness; Clarke et al(2006a), dispersion weighting; (2006c), resemblancesand sparsity; (2008), similarity profiles and linkagetrees; (2014), shade plots; and Somerfield and Clarke(2013), coherent species curves.ACKNOWLEDGEMENTSAny initiative spanning quite as long a period as thePRIMER software represents (the first recognisableelements were committed to paper over 30 years ago)is certain to have benefited from the contributions ofa vast number of individuals: colleagues, students,collaborators and a plethora of PRIMER users. Somuch so, that it would be invidious to try to produce alist of names – we would be certain to miss outimportant influences on the development of the ideasand examples of this manual and thereby lose goodfriends! But we are no less grateful to all who haveinteracted with us in connection with PRIMER andthe concepts that this manual represents. One namecannot be overlooked however, that of Prof MartiAnderson (Massey University, NZ); our collaborationwith Marti, in which her research has been integratedinto add-on software (PERMANOVA ) to PRIMER,has further broadened and deepened these concepts.Similar sentiments apply to funding sources: most ofthe earlier work was done while all authors wereemployed by Plymouth Marine Laboratory (PML),and for the last 14 years two of us (KRC, RNG) havemanaged to turn this research into a micro-business(PRIMER-E Ltd) which, though operating quiteindependently of PML, continues to have close ties toits staff and former staff, represented by the other twoauthors (PJS, RMW). We are grateful to the formersenior administrators in the PML and the NaturalEnvironment Research Council of the UK whoactively supported us in a new life for this research inthe private sector – it has certainly kept us out ofmischief for longer than we had originally expected!Prof K R Clarke (founder PRIMER-E and Hon Fellow, PML)R N Gorley (founder PRIMER-E)Dr P J Somerfield (PML)Prof R M Warwick (Hon Fellow, PML)2014

Chapter 1page 1–1CHAPTER 1: A FRAMEWORK FOR STUDYING CHANGES INCOMMUNITY STRUCTUREThe purpose of this opening chapter is twofold:a) to introduce some of the data sets which are usedextensively, as illustrations of techniques, throughout the manual;b) to outline a framework for the various possiblestages in a community analysis¶.or relative criteria (‘under impact, this coefficientis expected to decrease in comparison with controllevels’). Note the contrast with the previous stage,which is restricted to demonstrating differencesbetween groups of samples, not ascribing directional change (e.g. deleterious consequence).Stages4) Linking to environmental variables and examiningissues of causality of any changes. Having allowedthe biological information to ‘tell its own story’,any associated physical or chemical variablesmatched to the same set of samples can be examinedfor their own structure and its relation to the bioticpattern (its ‘explanatory power’). The extent towhich identified environmental differences areactually causal to observed community changescan only really be determined by manipulativeexperiments, either in the field or through laboratory/mesocosm studies.It is convenient to categorise possible analyses broadlyinto four main stages.TechniquesExamples are given of some core elements of therecommended approaches, foreshadowing the analysesexplained in detail later and referring forward to therelevant chapters. Though, at this stage, the detailsare likely to remain mystifying, the intention is thatthis opening chapter should give the reader some feelfor where the various techniques are leading and howthey slot together. As such, it is intended to serveboth as an introduction and a summary.1) Representing communities by graphical descriptionof the relationships between the biota in the varioussamples. This is thought of as pure description,rather than explanation or testing, and the emphasisis on reducing the complexity of the multivariateinformation in typical species/samples matrices, toobtain some form of low-dimensional picture ofhow the biological samples interrelate.2) Discriminating sites/conditions on the basis of theirbiotic composition. The paradigm here is that ofthe hypothesis test, examining whether there are‘proven’ community differences between groups ofsamples identified a priori, for example demonstrating differences between control and putativelyimpacted sites, establishing before/after impactdifferences at a single site, etc. A different type oftest is required for groups identified a posteriori.3) Determining levels of stress or disturbance, byattempting to construct biological measures from thecommunity data which are indicative of disturbedconditions. These may be absolute measures (“thisobserved structural feature is indicative of pollution”)¶The term community is used throughout the manual, somewhatloosely, to refer to any assemblage data (samples leading to counts,biomass, % cover, etc. for a range of species); the usage does notnecessarily imply internal structuring of the species composition,for example by competitive interactions.The spread of methods for extracting workable representations and summaries of the biological data can begrouped into three categories.1) Univariate methods collapse the full set of speciescounts for a sample into a single coefficient, forexample a species diversity index. This might besome measure of the numbers of different species(species richness), perhaps for a given number ofindividuals, or the extent to which the communitycounts are dominated by a small number of species(dominance/evenness index), or some combinationof these. Also included are biodiversity indices thatmeasure the degree to which species or organismsin a sample are taxonomically or phylogeneticallyrelated to each other. Clearly, the a priori selectionof a single taxon as an indicator species, amenableto specific inferences about its response to a particular environmental gradient, also gives rise to aunivariate analysis.2) Distributional techniques, also termed graphicalor curvilinear plots (when they are not strictlydistributional), are a class of methods whichsummarise the set of species counts for a singlesample by a curve or histogram. One example is kdominance curves (Lambshead et al, 1983), whichrank the species in decreasing order of abundance,convert the values to percentage abundance relativeto the total number of individuals in the sample,

Chapter 1page 1–2and plot the cumulated percentages against thespecies rank. This, and the analogous plot basedon species biomass, are superimposed to defineABC (abundance-biomass comparison) curves(Warwick, 1986), which have proved a useful construct in investigating disturbance effects. Anotherexample is the species abundance distribution(sometimes termed SAD curves or the distributionof individuals amongst species), in which thespecies are categorised into geometrically-scaledabundance classes and a histogram plotted of thenumber of species falling in each abundance range(e.g. Gray and Pearson, 1982). It is then argued,again from empirical evidence, that there arecertain characteristic changes in this distributionassociated with community disturbance.or an ordination plot in which, for example, thesamples are ‘mapped’ (usually in two or threedimensions) in such a way that the distancesbetween pairs of samples reflect their relativedissimilarity of species composition.Methods of this type in the manual include: hierarchical agglomerative clustering (see Everitt, 1980) inwhich samples are successively fused into largergroups; binary divisive clustering, in which groupsare successively split; and two types of ordinationmethod, principal components analysis (PCA, e.g.Chatfield and Collins, 1980) and non-metric/metricmulti-dimensional scaling (nMDS/mMDS, the formeroften shortened to MDS, Kruskal and Wish, 1978).For each broad category of analysis, the techniquesappropriate to each stage are now discussed, andpointers given to the relevant chapters.Such distributional techniques relax the constraintin the previous category that the summary fromeach sample should be a single variable; here theemphasis is more on diversity curves than singlediversity indices, but note that both these categoriesshare the property that comparisons between samples are not based on particular species identities:two samples can have exactly the same diversity ordistributional structure without possessing a singlespecies in common.UNIVARIATE TECHNIQUESFor diversity indices and other single-variableextractions from the data matrix, standard statisticalmethods are usually applicable and the reader isreferred to one of the many excellent generalstatistics texts (e.g. Sokal and Rohlf, 1981). Therequisite techniques for each stage are summarised inTable 1.1. For example, when samples have thestructure of a number of replicates taken at each of anumber of sites (or times, or conditions), computingthe means and 95% confidence intervals gives anappropriate representation of the Shannon diversity(say) at each site, with discrimination between sitesbeing demonstrated by one-way analysis of variance(ANOVA), which is a test of the null hypothesis thatthere are no differences in mean diversity between3) Multivariate methods are characterised by the factthat they base their comparisons of two (or more)samples on the extent to which these samples shareparticular species, at comparable levels of abundance. Either explicitly or implicitly, all multivariatetechniques are founded on such similarity coefficients, calculated between every pair of samples.These then facilitate a classification or clustering¶of samples into groups which are mutually similar,Table 1.1. Univariate techniques. Summary of analyses for the four stages.Univariate examplesStages1) Representingcommunities2) Discriminatingsites/conditionsDiversity indices (Ch 8)Indicator taxaBiodiversity indices (Ch 17)Means and 95% confidence intervals for each site/condition (Ch 8, 9, 17)1-way analysis of variance, ANOVA, Ch 6 (collectively, multivariate tests can be used, Ch 6)3) Determiningstress levelsBy reference to historical data for sites (Ch 14, 15) and regional ‘species pool’ (Ch 17)Ultimately a decrease in diversity Initial increase in opportunists Loss of taxonomic distinctness4) Linking toenvironmentRegression techniques, Ch 11 (collectively, BEST, Ch 11); for causality issues see Ch 12¶These terms tend to be used interchangeably by ecologists, so we will do that also, but in statistical language the methods given here areall clustering techniques, classification usually being reserved for classifying unknown new samples into known prior group structures.

Chapter 1page 1–3sites. Linking to the environment is then also relatively straightforward, particularly if the environmentalvariables can be condensed into one (or a small numberof) key summary statistics. Simple or multiple regression of Shannon diversity as the depend

package PRIMER (Plymouth Routines In Multivariate Ecological Research), obtainable from PRIMER-e, (see Its scope is the analysis of data arising in community ecology and environmental science which is multivariate in character (m