Cambridge Books Online

http://ebooks.cambridge.org/

Multivariate Analysis of Ecological Data using CANOCO 5

Petr milauer, Jan Lep
Book DOI: http://dx.doi.org/10.1017/CBO9781139627061
Online ISBN: 9781139627061
Paperback ISBN: 9781107694408

Chapter
2 - Using Canoco 5 pp. 15-38
Chapter DOI: http://dx.doi.org/10.1017/CBO9781139627061.003
Cambridge University Press

Using Canoco 5

In this chapter you will learn about the style of work which Canoco 5 supports and
encourages and about the basic techniques of importing and editing data tables, defining
and executing statistical analyses, and visualising their results. This introduction to the
use of Canoco 5 is necessarily brief and cannot cover all its aspects. You are advised
to obtain additional information from the Canoco 5 manual, starting with its tutorial
(Chapter 2).

2.1

Philosophy of Canoco 5
Canoco 5 is implemented around a way of working with data, statistical analyses and
their results, which is quite different from earlier versions. All the work with data and
analyses is now more tightly integrated and when properly understood, Canoco 5 aids
users not only in routine work, but also in making correct decisions about the statistical
analyses. The relations among the most important entities of work in Canoco 5 are
summarised in Figure 21.
The central stage is taken by projects and analyses. A project contains one or multiple
data tables, as well as analyses of the data. You can open only one project in the
Canoco 5 program at any time. But you can also open multiple Canoco 5 instances
(by repeatedly double-clicking the Canoco 5 icon at the computer desktop) and work
with different projects in each. Each analysis in a project represents a single task,
addressing one or multiple research questions you have about the data. When an analysis
is accomplished (by choosing an appropriate analysis template, specifying its options,
and executing the analysis), analysis results can be inspected in an analysis notebook,
which is automatically opened after the analysis is performed and contains one to
many pages. When working with ordination methods, visual representation of their
results plays a key role and Canoco 5 makes it easy to quickly create appropriate
ordination diagrams using the Graph Wizard. With the help of commands located in
the Graph menu a large variety of additional ordination diagrams and attribute plots
can be created. All created graphs are also contained, as separate pages, in the analysis
notebook.

Downloaded from Cambridge Books Online by IP 128.122.253.228 on Thu Apr 30 03:45:05 BST 2015.
http://dx.doi.org/10.1017/CBO9781139627061.003
Cambridge Books Online Cambridge University Press, 2015

16

Using Canoco 5

Project
Data tables
table 1

Analyses
analysis 1

analysis 2

Notebooks

notebook 1

notebook 2

table 2
analysis 3

notebook 3

Figure 21 Canoco 5 project components.

Canoco Adviser and the Wizards

Canoco Adviser is a mysterious character of the Canoco 5 user interface: you read
a lot about him in the software documentation, but you never meet him in person. He
works with you through multiple wizards. Wizards are specific sequences of dialog
pages, which help you to accomplish complex tasks by dividing them into simpler
sub-tasks on individual pages. You proceed through a page sequence using the Next
button, until you come to the end, where the Next button is replaced by another
one, labelled Finish. In addition, each wizard page has a Back button to return to an
earlier page and a Help button, displaying a help window explaining the contents of
the current page.
The page sequence shown in each wizard is not fully fixed, the number and identity
of displayed pages depends on the context. Beside displaying a preset range of options,
wizard pages sometimes provide advice based on the type and statistical properties
of the analysed data or on the type of chosen analysis. Such advice is assembled by an
expert system built into Canoco 5. The entire set of such advisory features is named
the Canoco Adviser.
Import Wizards help you select the data present in Excel spreadsheet files or in
legacy data files formatted for older versions of CANOCO, describe their properties,
and transform them into project data tables.

Downloaded from Cambridge Books Online by IP 128.122.253.228 on Thu Apr 30 03:45:05 BST 2015.
http://dx.doi.org/10.1017/CBO9781139627061.003
Cambridge Books Online Cambridge University Press, 2015

2.2 Data import and editing

17

New Analysis Wizard suggests various types of analyses (analysis templates) that
are appropriate for the data available in your project and allows you to select one of
them, matching your task. Based on your choice, a new analysis is then created. New
analyses can be created also without the wizard as so-called customised analyses
(including imported CANOCO 4.x project files), but this is not recommended if you
can achieve your task using the wizard.
Analysis Setup Wizard takes a newly created or existing analysis and allows you
to adjust its settings (variables to use, number of ordination axes to compute, the type
of test to apply, etc.). It corresponds to the only wizard present in CANOCO 4.5.
Graph Wizard is available for most analyses created with the New Analysis
Wizard and it is shown automatically the first time the analysis was executed. It
provides quick access to a subset of graphing utilities of Canoco 5.

2.2

Data import and editing

The preparation of input data for multivariate analyses has always been one of the big
obstacles to their effective use. In earlier versions of the CANOCO program, one had
to understand the overly complicated and unforgiving format of the data files, which
was based on the requirements of the FORTRAN programming language on which
CANOCO was based. Version 4.x of CANOCO alleviated this problem by an easy
method of transforming data stored in spreadsheets into CANOCO format files via the
export utility called WCanoImp.
Canoco 5 now offers direct import of Excel spreadsheet files into Canoco project.
Alternatively, you can import in a similar way the legacy CANOCO 4.x data files or
you can enter the data directly within Canoco 5 workspace, using its integrated spreadsheet editor. This introduction describes only the import from Excel spreadsheet files.
Let us start with the data in your spreadsheet files. While the majority of users will work
with Microsoft Excel, Excel formats can be also exported from open-source spreadsheet
software running under Microsoft Windows. If the data are stored in a relational database
(Oracle, FoxBASE, Access, etc.), you can use the facilities of your spreadsheet program
to first import the data into it. In the spreadsheet, you must arrange your data into
a rectangular structure, as laid out by the spreadsheet grid. In the default layout, the
individual cases correspond to the rows, while the individual spreadsheet columns
represent the variables.1
In addition, you should have at least a simple heading for both rows and columns: the
first row (except the empty upper left corner cell) contains the names of variables, while
the first column contains the names of individual cases. Unlike the earlier versions of
CANOCO 4.x, the length of the row and column names is not too limited (it should not
exceed 255 characters). Nevertheless, brief names as defined in the earlier CANOCO
1

This is the layout always used in data tables of Canoco 5 projects, but Excel spreadsheets to be imported
may have this arrangement rotated by 90 degrees.

Downloaded from Cambridge Books Online by IP 128.122.253.228 on Thu Apr 30 03:45:05 BST 2015.
http://dx.doi.org/10.1017/CBO9781139627061.003
Cambridge Books Online Cambridge University Press, 2015

18

Using Canoco 5

Figure 22 First page of the Excel Import Wizard after one input file was added.

versions are still very useful for creating informative ordination diagrams. Canoco
5 therefore generates them from the full names while importing the data. Alternatively,
you can have the first two rows of the source spreadsheet filled with full and brief names
(in this order) of columns and/or the first two columns filled with full and brief names
of rows (cases). The brief names cannot have more than eight characters. When you let
Canoco 5 generate the brief names during import, you can still adjust their contents later
in the Canoco 5 spreadsheet editor.2
The remaining cells of the spreadsheet must be numbers (whole or decimal) for
numeric variables or text values for factors or they can be empty. An empty cell in
imported data always implied a zero value in CANOCO 4.x, but in Canoco 5, empty cells
can be used to code either zero values or the missing values and this choice can be set for
each data table before its import.3 Missing values can be always specified using NA text.
Data can be imported from an Excel spreadsheet file either as a part of creating a new
Canoco 5 project (using the File | Import project | from Excel menu command) or you
can add one or more tables to an existing project (using the Data | Add new table(s) |
2

Note that the Canoco 5 spreadsheet shows, at any time, only the brief or only the full names and you
can change the current display (separately for table rows and columns) e.g. using the Data | Short Labels
submenu.
For a factor variable, empty cell always means a missing value, because zero value makes sense only for
numeric variables.

Downloaded from Cambridge Books Online by IP 128.122.253.228 on Thu Apr 30 03:45:05 BST 2015.
http://dx.doi.org/10.1017/CBO9781139627061.003
Cambridge Books Online Cambridge University Press, 2015

2.2 Data import and editing

19

Figure 23 Second page of the Excel Import Wizard when only one input file was specified on

preceding page.

Import from Excel menu command). Either way, the user interface of the Excel import
wizard is essentially identical and we describe it briefly here.
The first page of the wizard (as illustrated in Figure 22) allows you to choose one
or multiple Excel files containing your data. Canoco 5 is able to import either from .xls
files with the format used up to Microsoft Office 2003 or from .xlsx files used in the
newer Office versions.4
On the following wizard page (illustrated in Figure 23), you must identify the sheets
containing the data and also specify how many tables shall be created. Note that this
count does not follow from the number of chosen sheets, because one data table can
be created by combining multiple rectangular areas possibly distributed across multiple
sheets, and one sheet can also contain multiple separate data tables (as in the case
illustrated here).
The number of data tables specified at this wizard page is then reflected in the following
wizard pages: for each table, the wizard displays two consecutive pages: the first page
in each pair is used to set table options, while the second one is used to specify the exact
range(s) of spreadsheet cells containing the data to be imported.
4

These two Excel formats cannot be combined in the same import wizard session, but this limitation can be
circumvented with the ability of Canoco 5 to add tables to an existing project.

Downloaded from Cambridge Books Online by IP 128.122.253.228 on Thu Apr 30 03:45:05 BST 2015.
http://dx.doi.org/10.1017/CBO9781139627061.003
Cambridge Books Online Cambridge University Press, 2015

20

Using Canoco 5

Figure 24 First of the two pages defining particular data table in the Excel Import Wizard.

Figure 24 illustrates the first of the two pages as shown for the first imported data
table. Details for individual options can be found for example by clicking the Help
button, but here we emphasise the necessity to specify carefully the terms to be used for
each tables rows and columns. For both kinds of terms, you must specify their singular
and plural. Also, the interpretation of empty cells is set here, in the lower part of the
page, and the table content is characterised as compositional data in the snapshot (see
the explanation at the end of Section 1.2).
It is advisable to import the table containing response data as the first data table in
the project. This will make your work with New Analysis Wizard easier, and the first
imported data table is also specified as compositional type by default, corresponding
to the fact that in many applications, the response data describe community composition, often explained by the environmental properties that would then naturally
come in a second data table of the project.
You should never merge the response and explanatory variables into a single
data table, always keep the response data as a separate table!

Downloaded from Cambridge Books Online by IP 128.122.253.228 on Thu Apr 30 03:45:05 BST 2015.
http://dx.doi.org/10.1017/CBO9781139627061.003
Cambridge Books Online Cambridge University Press, 2015

2.2 Data import and editing

21

Figure 25 Second of the two pages defining particular data table in the Excel Import Wizard.

For the second and later data tables of a project, the terms for their rows cannot be
specified ad hoc. Instead, they must be chosen from the terms used for rows (or columns)
of the tables defined earlier. This simple constraint guarantees coherence among multiple
tables of a project and this coherence is also reflected in the synchronisation of the number
of rows (cases) across project tables. When you insert or delete rows in one of the tables,
the same number of rows is inserted or removed in corresponding positions of the other
tables sharing the same row identity. In many projects, row identity is the same for all
data tables, but another possibility is to base the identity of the rows of a table on the
columns of an earlier defined data table. Such a table can then represent, for example,
species traits for the species defined by columns of an earlier data table.
The second of the two wizard pages shown for each imported data table (illustrated in
Figure 25) displays in its central part a preview of the content of the sheet(s) selected
on the preceding page. The Excel import wizard attempts to guess the position of the
rectangle with input data, as well as the presence and type of row and column labels.
Both options can be adjusted, however.
Downloaded from Cambridge Books Online by IP 128.122.253.228 on Thu Apr 30 03:45:05 BST 2015.
http://dx.doi.org/10.1017/CBO9781139627061.003
Cambridge Books Online Cambridge University Press, 2015

22

Using Canoco 5

Figure 26 Canoco 5 workspace with integrated spreadsheet window in the foreground.

After you have specified import options for all data tables, the wizard imports the data
into a new (or existing) Canoco 5 project and it is shown in a Canoco 5 spreadsheet,5
illustrated in Figure 26.
This spreadsheet looks similar to the Microsoft Excel user interface and you can
indeed enter new (or change existing) values there. But there are also differences: the
Canoco 5 spreadsheet offers some functionality specific for working with multivariate
data (e.g. recoding numeric variables into factors and vice versa) and some of the Excel
functionality (such as using formulas) is not available. The labels are treated specifically:
you can always see just one of the two label sets (full or brief labels, see above) and
they always stay at the left and top edges of the window, even when you scroll the
5

Before that, you might also see a dialog box offering Introductory Analysis and if you select the Yes button
in it, analysis setup is offered. After you execute the analysis, its notebook comes into foreground. But
the data spreadsheet is still available and can be displayed e.g. using the Data | Show project data menu
command.

Downloaded from Cambridge Books Online by IP 128.122.253.228 on Thu Apr 30 03:45:05 BST 2015.
http://dx.doi.org/10.1017/CBO9781139627061.003
Cambridge Books Online Cambridge University Press, 2015

2.2 Data import and editing

23

Figure 27 Canoco workspace with visible data table spreadsheet containing three numeric

variables and one factor (Management).

spreadsheet contents. Each data table is represented by one page in the spreadsheet
window and you can choose the currently visible table using the tabs shown at the top
of the spreadsheet.
You can create a new project also by importing data from files used by older CANOCO
versions (4.0 and 4.5) using the File | Import project | from Canoco 4.x files menu
command. The corresponding import wizard is simpler, as the individual imported files
match the newly created data tables one-to-one. Using this wizard, you can also import
CANOCO 4.x project files (with .con file extension), turning them into analyses in
Canoco 5 project. After importing legacy data containing factor predictors, you must
turn the dummy variables (used in older versions to code factor values) into real factor
columns, as described below.
After you create a new project by import or even after you defined some analyses
in the project, you can still add new data tables to it. This is accomplished through
the commands in the Data | Add new table(s) submenu. These commands allow you to
import additional data tables from Excel or CANOCO 4.x data files or to create empty
ones or, to support advanced analyses, to define data tables in a specialised way, e.g.
as summary statistics of other data tables or as averages of functional trait values per
individual cases, or as a distance matrix computed from a phylogenetic tree.
You can also create a new Canoco 5 project with empty data tables and gradually
fill them by entering data directly in the spreadsheet visible in Canoco workspace. The
spreadsheet supports automatic data table growth, as you enter new values into its border
zone (data cells with light orange background colour) at the bottom and right edges of
the spreadsheet.
Variables present in Canoco 5 data tables can be either numeric ones or factors,
which are (under default settings) recognisable by their light blue background colour, as
illustrated in Figure 27.6
6

Semi-quantitative variables, particularly those on ordinal scale, are not treated in any special way in Canoco
and we recommend using them as numeric ones, with necessarily more cautious interpretation of the results.

Downloaded from Cambridge Books Online by IP 128.122.253.228 on Thu Apr 30 03:45:05 BST 2015.
http://dx.doi.org/10.1017/CBO9781139627061.003
Cambridge Books Online Cambridge University Press, 2015

24

Using Canoco 5

In CANOCO 4.x versions, each factor had to be coded using so-called dummy
variables. There was one separate dummy variable for each different value (level)
of a factor. If a case (observation) had a particular value of the factor, then the
corresponding dummy variable had the value 1.0 for this case and the other dummy
variables had zero values for the same case. This explicit coding of factors by dummy
variables is no longer used in Canoco 5, but it is still performed internally (as in the
other statistical packages using factors), and it is reflected in ordination diagrams,
where each factor level is represented by a separate symbol.7
But the explicit decomposition of factors into dummy variables is still required, when
you want to create so-called fuzzy coding. As an example, you might be recording
vegetation and you record the type of agricultural management at each site, e.g. as a
factor with three levels: pasture, meadow, and abandoned. To record such data in Canoco
5, a single column with a factor variable is sufficient. But imagine that you might have
in your data set a site that had been used as a hay-cut meadow until the previous year, but
it is used as a pasture in the current year. You can reasonably expect that both types of
management influenced the present community composition. Therefore, you can define
three dummy (fuzzy-coding) variables named, say, Meadow, Pasture and Abandoned
and for such a mixed case give values larger than 0.0 and less than 1.0 for both
the first and second variable. The important restriction here is that the values of all the
variables coding such a fuzzy factor must sum to 1.0. Unless you can quantify the relative
importance of the two management types acting on this site, your best guess is to use
values 0.5, 0.5, and 0.0 (respectively for the Meadow, Pasture, and Abandoned variables).

2.3

Defining analyses
With data present in a Canoco 5 project, you can start addressing your research questions
using analyses. There are multiple ways of defining an analysis and here we focus on
their creation with the New Analysis Wizard. It can be started using the New . . . button
below the list of analyses (in the lower left corner of the Canoco 5 workspace). This
wizard is relatively simple, with one to three pages (the first two pages are not shown if
your project contains just a single data table). On the first page, all data tables present
in your project are initially selected, but if you do not want to use some of them in the
new analysis, uncheck their boxes. The second page displays selected data tables and
asks you to choose the focal table for your analysis. Typically, this is the table with the
response data that you try to explain using the ordination model.8 Most important is the
last page of the New Analysis Wizard, illustrated in Figure 28.
7

Canoco 5 supports replacement of a set of dummy variables (e.g. imported from a CANOCO 4.x data file)
with a single factor variable. To do so, select all the dummy variables forming the factor, right-click the
header of the first variable and select the Aggregate dummy variables menu command. Reverse change is
also possible by selecting one factor variable, right-clicking its header and choosing the Expand into dummy
variables command.
In more advanced analyses relating functional traits and environmental descriptors, the choice made on
this page is less trivial, as described for corresponding analysis templates in the Canoco 5 manual (section
4.3.4.5).
Downloaded from Cambridge Books Online by IP 128.122.253.228 on Thu Apr 30 03:45:05 BST 2015.
http://dx.doi.org/10.1017/CBO9781139627061.003
Cambridge Books Online Cambridge University Press, 2015

2.3 Defining analyses

25

Figure 28 Last page of New Analysis Wizard, with suggested analysis templates.

This page offers multiple analysis templates, arranged into a set of categories (represented by the bold text in the largest dialog field). Categories like Variation Partitioning
Analyses in Figure 28 are not empty, the list of their members can be seen when you
unfold the category by double-clicking its title.
Each analysis template represents a specific type of analytical task, with its roles
assigned to data tables in your project. In the lower part of the dialog, a text field
describes (using project-specific terms plant species and environmental variables in
the above figure) what you can achieve with the particular template. Standard ordination
analyses of constrained and unconstrained ordination are offered in the first Standard
Analyses category and the Canoco Adviser initially suggest one of these templates for
you (depending on the state of your project).
After you select an analysis template and click the Finish button, the new analysis is
created in your project based on that template and it is placed into the list of analyses.
But before you can execute the analysis, its settings must be adjusted or at least reviewed.
Therefore, another wizard window is shown, called the Analysis Setup Wizard.9

You can also display this wizard for already defined (and even executed) analysis by selecting the analysis
and clicking the Modify button below the list of analyses. Canoco 5 gives you a choice of either modifying
the existing analysis or first creating its copy and making the changes to it.

Downloaded from Cambridge Books Online by IP 128.122.253.228 on Thu Apr 30 03:45:05 BST 2015.
http://dx.doi.org/10.1017/CBO9781139627061.003
Cambridge Books Online Cambridge University Press, 2015

26

Using Canoco 5

Figure 29 QuickWizard mode is active: the QuickWizard toggle button, fifth from the left in the

toolbar, looks pushed-in.

Quick Wizard and Slow Wizard

The Analysis Setup Wizard can present itself in two modes: for routine analyses,
where you do not need any fine-tuning of the options, the so-called QuickWizard
mode is an appropriate choice. It is, in fact, a default setting after you install the
software. With QuickWizard mode set, the wizard shows just two pages for a plain
unconstrained ordination (PCA, DCA or CA). If you ask for unconstrained ordination with supplementary variables, one page is added for supplementary variables
selection. For a standard constrained ordination (RDA or CCA), you will see one
(or more) additional page(s) for specifying permutation test options, beside the page
where explanatory variables can be chosen.
If you need, however, to select a subset of cases or subset of response variables or
if you want to perform explicit weighting of particular cases or to create interaction
terms for your predictors, you must switch to Slow Wizard mode, before you define
the new analysis. To do so, click the QuickWizard button in the Canoco 5 main
toolbar (illustrated in Figure 29). It is a toggle button to restore the QuickWizard
mode, click it again.
Detailed information about the options available in setup wizard pages can be found
in the Canoco 5 manual (pp. 111143) and the work with this wizard is also illustrated by
some of the tutorials in this book. Here we will discuss just a single page, named Ordination Options and representing the core setting page for unconstrained and constrained
ordination. Its look (for a compositional response data) is illustrated in Figure 210.
At the page top, four main types of ordination methods supported by Canoco 5
(see Section 4.7) are shown.10 Partial analyses (with covariates) or analyses with supplementary variables are not explicitly distinguished here, just by the presence of the
corresponding variable type in the analysis. Similarly the hybrid constrained analyses,
explicitly identified in earlier CANOCO versions, are now distinguished by a nondefault choice in the Hybrid analysis field in the middle of the page. The two choices for
10

Specialised multivariate methods newly available in Canoco 5 (e.g. principal coordinates analysis, nonmetric multidimensional scaling, or co-correspondence analysis) do not show this page in their Analysis
Setup Wizard sequence.

Downloaded from Cambridge Books Online by IP 128.122.253.228 on Thu Apr 30 03:45:05 BST 2015.
http://dx.doi.org/10.1017/CBO9781139627061.003
Cambridge Books Online Cambridge University Press, 2015

2.3 Defining analyses

27

Figure 210 Ordination Options page of the Analysis Setup Wizard.

unconstrained ordination are disabled in the above snapshot, as the use of constrained
ordination was fixed in the New Analysis Wizard by selecting the Constrained analysis
template.
Canoco 5 allows you to choose between the linear and the unimodal ordination
model,11 but the choice shown for your analysis is not a dumb default (as it used to be
in earlier CANOCO versions). Instead, the Canoco Adviser performed, in a background
check, an unconstrained unimodal ordination (DCA) and measured the length of its
ordination axes (see the report shown right below the analysis choice). In a DCA with
detrending by segments, this length is measured in so-called turnover (or SD) units and
for community composition data represents a measure of its beta diversity. The Canoco
Adviser applies here a traditional heuristic rule and for this data set (with the longest axis
of 3.7 turnover units) recommends a unimodal method, i.e. canonical correspondence
analysis (CCA). For data with this length smaller than three turnover units, the linear
method is offered as a preferable choice, while for a length of more than four turnover
units, the linear method is no longer recommended.12
11
12

Except when the response data table is marked as general, because the unimodal (weighted averaging)
ordination methods are appropriate only for compositional data.
Less technically, using an example of biotic community as your response data table, you can decide based
on your a priori (or field-based) knowledge of the extent of community change along the major gradients

Downloaded from Cambridge Books Online by IP 128.122.253.228 on Thu Apr 30 03:45:05 BST 2015.
http://dx.doi.org/10.1017/CBO9781139627061.003
Cambridge Books Online Cambridge University Press, 2015

28

Using Canoco 5

When deciding whether to use the linear or unimodal type of ordination, you must
take into account another important difference among them. The unimodal methods
always implicitly standardise the data. CCA, CA or DCA methods summarise the
variation in relative frequencies of response variables (e.g. species). An important
implication of this fact is that these methods cannot work with empty cases, i.e.
records in which no variable has non-zero value (e.g. no species is present). Also,
the unimodal ordination cannot be used when the response variables do not share
measurement units, but this mistake is now largely prevented by the Canoco Adviser.
On the other hand you can usefully apply unimodal ordination methods to data
sets with low compositional variation (where DCA would produce axes with small
length in turnover units), because they also have a linear face (see Canoco 5 manual,
p. 89).

The Computed axes field allows you to calculate more than the first four ordination
axes, but the default choice is sufficient for a great majority of real-world data sets.
The Detrending options are offered only for unimodal ordination methods. The
detrending of the second and higher ordination axes of a correspondence analysis (leading to detrended correspondence analysis, DCA) is often used to cope with the so-called
arch effect, illustrated in Figure 211 by a diagram with case positions on the first two
axes of correspondence analysis (CA).
The position of cases on the second (vertical) axis is here strongly but not linearly
dependent on their position on the first (horizontal) axis. This effect can be interpreted as
a limitation of the method, because the consecutive axes are made mutually independent
but only a linear independence is required or, alternatively, as a consequence of the
projection of the non-linear relations of response variables to the underlying gradients
into a linear Euclidean drawing space (see Legendre & Legendre 2012, pp. 482485 for
more detailed discussion). Detrending by segments (Hill & Gauch 1980), while lacking
a convincing theoretical basis and considered as inappropriate by some authors (e.g.
Wartenberg et al. 1987 or Knox 1989), is the most frequently used way to making
the recovered compositional gradient straight (linear). When you select a unimodal
ordination model for your analysis in the Ordination Options page of the Analysis Setup
Wizard, Canoco offers the following choices for the Detrending field, as illustrated in
Figure 212.
Use of detrending by segments is not recommended for unimodal ordination methods where either covariates or explanatory variables are present (i.e. for partial and/or
constrained methods). In such cases, if the detrending is needed, detrending by polynomials is the recommended choice. You are advised to check the Canoco 5 manual
reflected in the data. If species mostly change only their proportions on these gradients, then the linear
approximation is appropriate and so are the linear techniques. If you expect qualitative changes in the
community composition (many species appearing and disappearing along the gradient), then unimodal
(weighted-averaging) techniques are a better choice.

Downloaded from Cambridge Books Online by IP 128.122.253.228 on Thu Apr 30 03:45:05 BST 2015.
http://dx.doi.org/10.1017/CBO9781139627061.003
Cambridge Books Online Cambridge University Press, 2015

29

CA Axis 2

2.3 Defining analyses

CA Axis 1

Figure 211 Scatter of cases from correspondence analysis, featuring so-called arch effect. You
can see the arch by rotating your book by 180 degrees.

Figure 212 Detrending method selection in the Analysis Setup Wizard.

(section 4.4.2.7) for more details on how to decide among the use of polynomials of
second, third, or fourth degree.
But the detrending procedure is often not needed for a constrained unimodal ordination, when analysing the same data table that required detrending for unconstrained
ordination. If an arch effect occurs in CCA, this is often a sign of some unnecessary
explanatory variables being present. For example, there may be two or more explanatory
variables in your set, strongly correlated (positively or negatively and not necessarily
in a linear way) with each other. If you retain only one variable from such a group
and remove also the variables that are not strongly correlated with the other ones, but
do not have an important relation with response data, the arch effect often disappears.
The selection of a subset of explanatory variables can be performed using the forward
selection of explanatory variables in Canoco 5 see Section 5.6.
In the Response data transformation and standardization section, present in the lower
part of the Ordination Options page, you can choose one of the standardisations discussed
in Section 1.3. But even here, the Canoco Adviser tries to help and automatically selects

Downloaded from Cambridge Books Online by IP 128.122.253.228 on Thu Apr 30 03:45:05 BST 2015.
http://dx.doi.org/10.1017/CBO9781139627061.003
Cambridge Books Online Cambridge University Press, 2015

30

Using Canoco 5

Figure 213 Centring and standardisation options in the Ordination Options page.

for you either no transformation or log transformation (with its two parameters properly
selected, see Section 1.3), depending on the range and distribution of data values in the
response data table. You will note that there are no centring or standardisation options
(discussed in Section 1.3) offered in the snapshot in Figure 210. This is because the
unimodal methods imply a double standardisation by both cases and response variables,
as explained in Section 4.5.
Centring and standardisation choices made for a linear ordination may substantially
change the outcome of the analysis. In fact, by varying the choice, you can address
different questions about your data (see the discussion in Section 15.3). The choices
available for centring and standardisation in linear ordination methods (see Figure 213)
are discussed in detail in Section 1.3. Here we provide only a brief summary of the
consequences of individual centring and standardisation options.
Centring by cases (referred to as landscape quadrats in Figure 213) results in
a zero average for each table row. Similarly, centring by response variable (named bird
species in the above figure) results in a zero average for each column. Centring by
response variables is obligatory for a constrained linear ordination (RDA, to which
Figure 213 refers) or for any partial linear ordination (i.e. where covariates are used).
Standardisation by response variables (or by cases) results in the norm of each column
(or each row) being equal to one. The norm is the square root of the sum of squares
of the column (row) values. If you apply both centring and standardisation, the centring
is done first.13 Therefore, after centring and standardising by response variables, the
columns represent variables with zero average and unit variance.
The Downweight rare <response-variables> option is available only for unimodal
methods and it targets the sensitivity of chi-square distances (implicitly used by these
methods) to (rare) species with a low total abundance (see Section 6.2.2). We recommend
to choose it particularly when working with a large, heterogeneous set of cases, with
many response variables (e.g. species) having only a few non-zero values (occurrences).
13

And if you centre/standardise both by rows and columns, the centring/standardisation is done first by rows,
then by columns.

Downloaded from Cambridge Books Online by IP 128.122.253.228 on Thu Apr 30 03:45:05 BST 2015.
http://dx.doi.org/10.1017/CBO9781139627061.003
Cambridge Books Online Cambridge University Press, 2015

2.3 Defining analyses

31

An alternative (but not fully equivalent) approach is the omission of such variables. See
Canoco 5 manual, p. 124, for additional details.
Whatever pages the setup wizard sequence contains, the last page is always the Finish
page. On it, there is only one option you can change, namely a check-box labelled
Execute this analysis after Finish.14 By default, it is always checked and this implies
analysis execution immediately after you click the Finish button.
After you executed the analysis (either by leaving the option, described in the preceding
paragraph, checked, or by clicking the Perform button below the list of analyses),
analysis results are shown in an analysis notebook.15 The analysis notebook almost
always contains a Summary page and the pages with individual graphs created for the
analysis, and depending on its presentation mode (see the Brief and non-brief view
of analysis notebooks box below) also additional pages, namely those with computed
ordination scores and analysis log. Figure 214 illustrates the content of Summary page
for a constrained ordination set up in our example above.
The upper part of the Summary page provides an overview of the size and type of the
data used in the analysis (including the real number of model degrees of freedom for
chosen explanatory variables), followed by a summary of the explained variation and
additional statistics for each of the computed ordination axes. Interpretation of these
values is illustrated in the first two case studies (Chapters 12 and 13) and full details
can be found in section 5.2 of the Canoco 5 manual. Finally, results of permutation tests
performed on the constrained ordination model are shown.
For other kinds of analyses, some of this information may be missing and other
information can be present, namely the results of stepwise selection of explanatory
variables or the evaluation of their simple and conditional effects.

Brief and non-brief view of analysis notebooks

In the same way you can switch on and off the comprehensive content of Analysis
Setup Wizard using the QuickWizard button, you can also decide whether the analysis
notebooks are shown in a brief (default) or non-brief mode. To do so, select the
Edit | Settings | Canoco5 options menu command and in the General page uncheck
or check the Show brief version of notebooks with analysis results box. The effect of
such change is not immediately visible in already opened notebooks: you must close
the notebook(s) using the Hide button in the lower left corner of Canoco 5 workspace
and then reopen it again using the same button, now with changed text Show. Note
also that when you close an analysis notebook, Canoco 5 usually switches to another
analysis (when it exists), so you must possibly activate the original analysis before
re-displaying its notebook.

14
15

This check-box control is not shown when you invoke the Analysis Setup Wizard on an already executed
analysis and do not change any of its settings.
For a newly created analysis, the notebook is shown automatically. Alternatively, you can display it by
selecting the analysis name in the list of analyses and clicking the Show button. When the notebook is
visible, text of the button changes to Hide, with corresponding change in functionality.

Downloaded from Cambridge Books Online by IP 128.122.253.228 on Thu Apr 30 03:45:05 BST 2015.
http://dx.doi.org/10.1017/CBO9781139627061.003
Cambridge Books Online Cambridge University Press, 2015

32

Using Canoco 5

Figure 214 Summary page of an analysis notebook for a constrained ordination.

Unlike earlier versions, Canoco 5 now offers even very complex tasks as a single
analysis. For example, variation partitioning for two groups of predictors did require in
CANOCO 4.x at least three separately executed analyses (called projects by then) plus
additional calculations on their results. In Canoco 5, this is now a single analysis (with
optional testing and stepwise selection for each group of predictors) that immediately
provides a variation partitioning table, including the results of permutation tests. To
perform the partitioning, however, Canoco 5 must still fit multiple (partial) constrained
ordination models, corresponding to individual executions of the ordination algorithm,
and these components of complex analysis are called analysis steps. With multiple steps
in an analysis, the Summary page of the analysis notebook always refers just to one of
the steps, but you can use the scroll control in the pages left-hand corner to review the
results of other steps, when needed. Standard ordination analyses have just a single step
and its existence is hidden from the user.

Downloaded from Cambridge Books Online by IP 128.122.253.228 on Thu Apr 30 03:45:05 BST 2015.
http://dx.doi.org/10.1017/CBO9781139627061.003
Cambridge Books Online Cambridge University Press, 2015

2.4 Visualising results

33

Figure 215 First page of the Graph Wizard for a constrained ordination.

2.4

Visualising results
In this section, we introduce the basic steps of graph creation in Canoco 5. The diagram types available in Canoco 5 and their interpretation are treated in more detail in
Chapter 11.
The results provided by project analyses can be visualised in two ways. The full
toolbox of visualisation methods is available from the Graph menu shown for an active
analysis with existing results. But for the analyses created using the New Analysis
Wizard, there is an easy and powerful way to create an initial set of diagrams to inspect
analysis results. After a newly created analysis is executed, Canoco 5 displays the Graph
Wizard, where a set of graphs appropriate for the analysis template is offered.
In our example (see Figure 215), Graph Wizard suggests creating a biplot with
response variables (plant species) and explanatory variables (environmental variables),
as well as an attribute plot illustrating the change of plant species diversity across the
ordination space defined by the environmental descriptors. In this first page, you should
select the graph types you are interested in, and for the types you select, Graph Wizard
displays additional pages, where the graph content can be further fine-tuned, as illustrated
in Figures 216 and 217.
Figure 216 shows a Graph Wizard page displayed for the above-selected biplot of
plant species and environmental variables.
Some of the options are disabled, as they are not appropriate for current analysis or
graph. The Graph Wizard considers both plotted components (plant species scores and
the scores of environmental variables) as required contents, so they cannot be removed
from the plot. For response variables (here plant species), however, you can select only

Downloaded from Cambridge Books Online by IP 128.122.253.228 on Thu Apr 30 03:45:05 BST 2015.
http://dx.doi.org/10.1017/CBO9781139627061.003
Cambridge Books Online Cambridge University Press, 2015

34

Using Canoco 5

Figure 216 Graph Wizard page for standard ordination diagrams.

a subset of their total number, those best predicted (fitted) by the constrained ordination
axes. Another option available in this page is the possibility to create this ordination
diagram not only with the two most important ordination axes (first and second), but
also with other ones, for example the first and third axis.
Figure 217 shows a wizard page that allows easy creation of another useful graph,
displaying community diversity in its relation to environmental variables.
As in the earlier page, case scores and environmental variables are obligatory parts
of the diagram and cannot be avoided, although the positions of individual cases will
not be plotted in the diagram, if you choose the contour plot option in the page line
introduced by the and presented using a label. In such case, a loess smoother is fitted to
the original data and the resulting surface presented with isolines. Alternatively, symbol
plot displays case symbols with varying diameter, proportional to the diversity value of
a particular case, and the colour-coded plot displays symbols for all cases with the same
size, but with the fill colour varying along a colour scale (gradient) reflecting the case
diversity. Another important choice concerns the type of diversity measure visualised in
the diagram. Standard species richness (number of present species) is the default choice,
but other diversity measures, such as ShannonWiener H or N2 diversity index (inverse
of the Simpson index), are also available.
The Graph Wizard also offers other kinds of diagrams, not present in our example
analysis two types of diagrams useful when the explanatory variables represent a
classification (i.e. a single factor), a specialised diagram for principal response curves
(PRC) method (see Section 16.6 for a practical example), as well as a specialised
diagram for co-correspondence analysis (see Section 13.5 for an example). Unlike the
other diagrams, the PRC and co-correspondence analysis diagrams cannot be created

Downloaded from Cambridge Books Online by IP 128.122.253.228 on Thu Apr 30 03:45:05 BST 2015.
http://dx.doi.org/10.1017/CBO9781139627061.003
Cambridge Books Online Cambridge University Press, 2015

2.4 Visualising results

35

Figure 217 Graph Wizard page for attribute plot visualising alpha diversity changes.

using the commands in the Graph menu, only with the Graph Wizard. The wizard can be
invoked at any time using either its button in the graphing toolbar or the Graph | Advise
on graphs menu command.
When working with the commands present in the Graph menus submenus (Scatterplots, Biplots, Triplots, and Attribute plots), you can create some graphs not available
through the Graph Wizard, but this flexibility comes at the price of larger complexity.
For example, in the case of multi-step analyses (see the end of Section 2.3 above), you
must select the step that calculates the ordination scores for graphing with the Analysis |
Select graphing step menu command.
When creating diagrams using the commands in the submenus of the Graph menu
(and to a limited extent also when working with the Graph Wizard), the look and content
of the created diagrams are affected by the options chosen at three different levels in the
Canoco 5 user interface:
1. At the level of Canoco 5 application, the choices made through the four commands
in the Edit | Settings submenu (most importantly the Graphing options and Visual
attributes) affect created graphs in all projects and analyses opened in Canoco 5.
Dialog boxes shown by the commands in the Settings submenu are described in
Canoco 5 manual, section 7.3.3.
2. At the level of a particular Canoco 5 project, the changes made through the commands
in the Project menu affect all graphs created (after the settings change) in the analyses

Downloaded from Cambridge Books Online by IP 128.122.253.228 on Thu Apr 30 03:45:05 BST 2015.
http://dx.doi.org/10.1017/CBO9781139627061.003
Cambridge Books Online Cambridge University Press, 2015

36

Using Canoco 5

of the currently open project. To store them across different Canoco 5 sessions, you
must save the project before closing the application. The most important changes
at this level affect the classification of data items (cases, variables), definitions of
groups and of spatial or temporal series, as well as the suppression of a subset of data
items.
3. For a particular analysis, you can change the graphing options using the Analysis |
Plot creation options menu command. This includes settings like the identity of
plotted axes or the selection of plotted data items based on the information provided
in the analysis results (such as the fit of response variables into the ordination space
illustrated earlier). You can find additional information about this important dialog
box in Canoco 5 manual, section 7.6.9.
To apply new settings to an already created graph, you must recreate it using either
the green double-arrow in the graphing toolbar or the Recreate graph command in the
Graph menu.
If you prefer to have the axes of an ordination diagram labelled with the name of
ordination method and axis number (or even with additional information), you can
achieve this easily by checking the Show transformation and axis labels dialog box
in the General page of the dialog shown by the Edit | Settings | Graphing options
menu command. When checked, Canoco 5 displays before a diagram is created a
dialog box where default axis labels are suggested and the label text can be adjusted.
For general XY(Z) diagrams, axis labels are plotted by default and the dialog offers
additional on-fly transformations of plotted variables.
You can save the current state of any Canoco 5 graph in a separate file with a .c5g
extension. To do so, the selected graph must be first copied from its analysis notebook
using the Graph | Copy to standalone graph menu command and then saved. Both the
standalone graphs and those included in the analysis notebooks can be printed directly
from the program, copied to other applications through the Windows Clipboard, or
exported in a variety of graph formats: BMP, PNG, JPEG, TIFF, Adobe Illustrator, PDF,
and Enhanced Metafile Format.

2.5

Beware, CANOCO 4.x users!

Long-term users of the older versions of Canoco software have the advantage of greater
expertise with the choice of ordination methods and interpretation of their results, but
there are also some dangerous waters for them, when they set sail with Canoco 5. Unlike
the novice users, they have working habits to unlearn, because otherwise these habits
will stand in their way towards an enjoyable experience with the new version. Here we
summarise the nine main differences of Canoco 5 from the previous version CANOCO
4.5 and suggest that users of older versions learn them by heart.

Downloaded from Cambridge Books Online by IP 128.122.253.228 on Thu Apr 30 03:45:05 BST 2015.
http://dx.doi.org/10.1017/CBO9781139627061.003
Cambridge Books Online Cambridge University Press, 2015

2.5 Beware, CANOCO 4.x users!

37

1. Make sure that for each data table created in a Canoco 5 project you specify correctly
whether it is of compositional or general type (see the end of Section 1.2, Section
2.2, and Canoco 5 manual, p. 48). When specified incorrectly, you will not be able
to use the full range of the Canoco Adviser functionality or to find required analysis
templates in the New Analysis Wizard. If you get it wrong during project definition,
you can always correct it later (for a currently selected data table) using the Data |
Change table kind to . . . menu command. Data table type affects which analysis
templates the Canoco Adviser offers and if your intended analysis is not among the
offered templates, then perhaps the data table type must be changed.
2. When your data layout fits the widespread pattern of having a single compositional
data table (typically describing community composition for a set of cases) and one
or multiple data tables with explanatory variables, make sure the compositional data
table is imported first into the project. In this way, the data type (compositional vs.
general, see above) is preset correctly in the import wizard and your use of the New
Analysis Wizard is more straightforward, as you can rely on its default settings.
3. In CANOCO 4.x, the response variables were always called species, your cases were
samples and your predictors environmental variables. This was perhaps a good
choice for community ecologists, but not for everyone. In Canoco 5, the choice of
proper item terms is in your hands and if you play along, the expert advice provided
by the Canoco Adviser during your work with Canoco 5 will be more understandable.
Do not rely on default item terms offered when you define (import) the data tables,
but choose the terms appropriate for your research field. You do not have cases,
but traps, quadrats, stations; perhaps you do not have just species but plant
species, or gene loci or land-use categories; and your environmental variables
might be in fact better called experimental treatments or socio-economic impacts.
These terms are used not only in the menu commands and in the Analysis Setup and
Graphing Wizards, but also in the Describe Graph Contents tool that aids you when
interpreting ordination diagrams.
4. Data tables referring to the same set of cases must have identical number of rows, with
matching identity: Canoco 5 now works with missing (NA) values, so if you do not
have environmental descriptors for a particular record of community composition, the
corresponding data table row must contain NA values. Canoco 5 spreadsheet editor
takes care of maintaining the consistency across data tables (e.g. the case labels are
shared), so just go along with it.
5. Dummy (0/1) variables are no longer the appropriate representation of factors. If you
import old data files, get rid of dummy variables (convert them to factors) as soon
as possible. Neither principal response curves (PRC), nor the linear discriminant
analysis (LDA) can be set up properly with dummy variables; these methods are only
offered if there are appropriate factors in the data tables to be analysed.
6. You no longer need to transform numeric predictors (explanatory and supplementary
variables or covariates) before you import them into Canoco 5 project. You can set the
transformation for individual predictors using the Data | Default transformation and
standardization menu command. In fact, when you choose this command, the Canoco

Downloaded from Cambridge Books Online by IP 128.122.253.228 on Thu Apr 30 03:45:05 BST 2015.
http://dx.doi.org/10.1017/CBO9781139627061.003
Cambridge Books Online Cambridge University Press, 2015

38

Using Canoco 5

Adviser may suggest using log transformation for some variables.16 Transformation
settings you specify in this dialog are not reflected in the data tables themselves (they
show the non-transformed data values), but they are applied to the variables when
you execute your analyses.
7. Canoco 5 data tables have both the traditional short labels for cases and variables as
well as full labels without practical limitation on their length. If possible, try to define
(import) primarily the full labels, the short ones will be automatically generated from
them and you can correct the result when needed (see section 2.9.3 of the Canoco
5 manual). Brief labels are also used in ordination diagrams by default, but this can
be changed for any kind of data items using the Project | Visibility and labelling menu
command.
8. The choice of scaling for ordination scores is no longer a part of the analysis setup. For
an executed analysis, you can change its scaling type on the fly, using the Edit scaling
options . . . tool in the main toolbar (sixth from the left, next to the QuickWizard mode
button). See Sections 11.1 and 11.2 for additional information on how the chosen
scaling affects the interpretation of ordination diagrams. For specialised analyses
requiring particular score scaling (such as principal response curves or discriminant
analysis), the scaling is fixed within the analysis by the Canoco Adviser.
9. You might eventually find that there is a possibility of creating new analyses using the
Analysis | Add new analysis | Customized menu command, which feels very similar
to the way the analyses were created in CANOCO 4.5. But we urge you to resist
this temptation and create your new analyses using the New . . . button below the
list of analyses. This is because the customised analyses are second-class citizens in
Canoco 5 projects and should be used only for very advanced tasks, not supported by
the analysis templates of the Canoco Adviser, or when you need to validate Canoco
5 analysis template results with those done earlier in CANOCO 4.5 (in this case, the
customised analyses result from importing CANOCO 4.5 projects).

16

You should always check their suitability for your variables, of course.

Downloaded from Cambridge Books Online by IP 128.122.253.228 on Thu Apr 30 03:45:05 BST 2015.
http://dx.doi.org/10.1017/CBO9781139627061.003
Cambridge Books Online Cambridge University Press, 2015