The Canoco for Windows module provides a standard Microsoft Windows® user interface to set-up an ordination analysis, to calculate the ordination and to view the results.
Context-sensitive online help is available at each step of the analysis. After the ordination has been calculated, you can inspect the results that are in a log-window, or you can immediately plot the ordination diagram by clicking the CanoDraw button in Canoco for Windows.
Canoco for Windows snapshot
Project Setup Wizard
In Canoco for Windows an ordination analysis is specified and modified with a Project Setup Wizard. This Wizard consists of a series of pages where you select the data sets to be analyzed and the ordination method with its options. With the Project Setup Wizard you can also specify Monte Carlo permutation tests to determine the statistical significance of your ordination model. The Project Setup Wizard further allows you to omit particular samples, species, environmental variables or covariables from the analysis, change weights of particular species or samples in the analysis, or to define interaction terms for the ordination model. The following snapshot illustrates the deletion of some environmental variables.
Delete environmental variables wizard page
This page is displayed if you indicated on an earlier page that you wish to delete some environmental variables. Note that
Canoco for Windows communicates with you in terms of the names of the variables- and not in terms of their code numbers, as the earlier Canoco versions did! After clicking the >> button at the top of the Project Setup Wizard page, the selected variables will be moved to the right-hand list and, consequently, omitted from the analysis.
Statistical modelling and forward selection
Canoco for Windows provides efficient ways to specify and test canonical ordination models. These models can be viewed as an extension of multiple regression / analysis of variance models, the extension being that canonical ordination models explain a multivariate response (typically the community composition) by a set of explanatory variables. Canoco for Windows allows for a flexible specification of the explanatory variables and interaction terms in these models.
In Canoco for Windows there can be two kinds of explanatory variables: environmental variables and covariables. Environmental variables are the explanatory variables of prime interest, whereas covariables are the explanatory variables that code for background variation. The ordination is adjusted for the effects of the covariables, just as the covariate in classical covariance analysis allows for the adjustment of the treatment means for pre-treatment differences between units.
In addition to flexible model specification, Canoco for Windows can perform Forward selection of environmental variables, with the option of testing the significance of the considered model term before it enters the model. The results of an automatic forward selection are summarized in two tables (snapshot).
Forward selection results
The table at the top, headed Marginal effects, lists the individual environmental variables in order of the variance they explain singly, i.e. when that particular variable is used as the only explanatory variable (lambda-1). In the example shown above, Moisture is the most influential variable, followed by A1 (the thickness of the A1 horizon) and NM (the indicator of Nature Management).
The table at the bottom, headed Conditional effects, shows the environmental variables in order of their inclusion in the model, together with the additional variance each variable explains at the time it was included (lambda-A) and its significance at that time (P-value). In the example shown above, Moisture is the variable that is selected first. With Moisture in the model, it is not A1 but NM that adds the most variance. Adding NM significantly improves the model (P=0.005). The third selected variable, HF, does not contribute significantly (P=0.085). The parsimonious model that best explains the species composition thus consists of Moisture and NM.
Both tables can be copied via the clipboard to your wordprocessor or spreadsheet program.
Permutation tests
With the Monte Carlo permutation tests in Canoco for Windows you can assess the statistical significance of the environmental variables, and of the first ordination axis of a canonical analysis.
A unique feature of Canoco for Windows is that it can test whether one or more environmental variables have an effect on the community composition after taking into account the effect of other variables. For this, you need to specify the other variables as covariables. This option allows you to answer questions such as:
- Does the management regime have an effect on the dune meadow vegetation after accounting for the fact that the meadows differ in moisture status and in thickness of the A1 soil horizon?
- Does nutrient pollution affect the species composition after taking into account the natural variation in salinity of the water?
- Does the putative impact in a Before-After Control-Impact study have a significant effect on the species community after accounting for natural variation between sites and times?
In the simplest Monte Carlo permutation test, the sampling units are shuffled completely at random. Many software packages implement only this simple permutation type. However, completely random permutation is not appropriate when your sampling units are correlated, because your data form a time series or have a spatial layout, or because your sampling design or experimental design shows additional structure. Canoco for Windows overcomes this limitation of permutation tests by implementing appropriate permutation types for data from
- regular time series
- equi-spaced linear transects
- rectangular spatial grids of samples
- split-plot designs and related balanced two-level designs, e.g.
- nested sampling designs (e.g.samples within estuaries)
- repeated measurements designs
- BACI (Before-After Control-Impact) designs
Each of these stuctures may be replicated. Each such replicate is specified as a block in Canoco for Windows.
The split-plot design framework in Canoco for Windows allows you to specify valid permutation tests for a variety of commonly used designs. The split-plot design is a hierarchical design with two levels of units: whole-plots containing split-plots. Split-plots are the lowest level sampling units, i.e. the samples in your data file. Examples are samples-within-regions, plots-within-stands, plots-along-transects, releves-within-time-series. The split-plot design framework allows you to specify permutations at both the whole-plot level and the split-plot level (see snapshot)
Split-plot design page
For example, to test for an effect of factors acting at the whole-plot level, select the Freely exchangeable option in the list at the left side and the No permutation option in the list at the right side. These options allow proper permutation testing in many related nested or crossed two-level designs. An important special case is the analysis of Before-After Control-Impact (BACI) design by permutation testing. In a BACI with two or more sites and multiple before/after times, the sites are specified as whole-plots and the individual sampling times as split-plots. If the sites lie on a transect (at various distances from the impact source) and if sampling is weekly, specify in the split-plot design page above, linear transects for whole-plots and time series for split-plots. If the impact levels could be randomly assigned to sites (as is possible in ecotoxicological experiments), select freely exchangeable for sites and time series for times.
Data size limits
Canoco for Windows (as well as the console version of Canoco 4.5) is able to analyze species (primary) data sets with up to 25,000 samples and 5,000 species. An additional requirement is that the number of nonzero values in the data matrix is not larger than 750,000. For the environmental data set (the explanatory variables of prime interest), the maximum number of the variables is 1000 (q) and the maximum number of the samples is again 25,000 (m). The additional requirement is that m * (q-8) is not larger than 500,000. For the covariable data set, the maximum number of covariables is 2000 (p), the maximum number of samples is 25,000 (m) and the maximum value of m*p is 1,000,000.
CanoDraw for Windows can handle analyses with data sets of any size.