Questions
- Is there an upgrade available for users of older versions?
- Does Canoco for Windows run on Windows 7 and other Windows platforms?
- How does Canoco deal with the stability issues reported by Oksanen & Minchin (1997)?
- Does Canoco for Windows provide NMDS?
- Is Canoco for Windows able to analyse my data set with 20,000 samples and 600 species?
- How does Canoco for Windows compare to PC-ORD 4.0?
- I bought Canoco at Scientia Software, which is no longer distributing Canoco for Windows. Who should I ask for technical support?
Answers
1. Is there an upgrade available for users of older versions?
Yes, you can upgrade from the older versions of Canoco. Please, contact the distributor from which you bought your original Canoco version to get more precise information on availability and pricing. See
How to order? for addresses.
2. Does Canoco for Windows run on Windows 7 and other Windows platforms?
The versions of Canoco 4.5 released before Windows 7 was released do not work with it, while versions 4.55 and 4.56 do. The owners of older versions could either contact Canoco distributor or to download the updater on their own from the following place:
http://www.canodraw.com/updt_456.exe
Please note that the updater - like the Canoco 4.5 installation itself - must be run from an account with administrative rights.
3. How does the new Canoco version treat the instability of ordination results, as reported by Oksanen & Minchin (1997)?
To quote from the new Canoco Reference Manual: "Oksanen and Minchin (1997) showed that CANOCO 3.12 suffered from the same type of instability." ... "By comparison with another algorithm to extract the ordination axes, they showed that the use of more stringent convergence criteria gives results that are acceptably stable. In line with their proposals, CANOCO 4 uses a maximum number of iterations of 999 and a tolerance of 10E-6, which is in-between their strict and superstrict tolerance criteria."
The bug in the original DECORANA (Hill, 1979) code for Detrended Correspondence Analysis has been studied and repaired. It turned out that the change suggested by Oksanen and Minchin (1997) did not make the smoothing routine order-invariant. Another change was required as well, as agreed upon by Oksanen & Minchin in subsequent correspondence (see http://cc.oulu.fi/~jarioksa/pages/isbug.htm and www.microcomputerpower.com/canoco315).
4. Does Canoco for Windows provide NMDS?
Yes and no.
Yes, because the separate Windows program WinKyst provides NMDS and works smoothly with Canoco for Windows to add environmental variables to the ordination diagram. WinKyst is freely available from www.canodraw.com.
No, because NMDS is not fully integrated in Canoco version 4.5x. However, Canoco includes Principal Coordinates Analysis (PCO), which is a metric form of multidimensional scaling.
The constrained form of PCO (the distance-based RDA) is also supported. In addition, Canoco provides a wide array of ordination methods, which operate directly on the data, instead of on derived similarity or dissimilarity measures (as NMDS does). The advantages of the methods in Canoco over NMDS are that:
- The resulting ordination readily provides information on the species level, i.e. information for the original variables being analysed.
- The ordination can be focused on the effect of particular explanatory variable(s) (direct gradient analysis) so that the ordination shows the community response to explanatory variables.
- Background variation can be removed from the ordination (by using covariable data) so that the ordination can display new information rather than trivial or uninteresting variation, or variation that is already well understood.
5. Is Canoco for Windows able to analyze my data set with 20,000 samples and 600 species?
It depends: while Canoco for Windows can analyze data sets of up to 25,000 samples and 5,000 species, there is a further limit: the number of nonzero values in the data matrix must be less than 750,000. So, if your species list has 600 items, but the average number of species per sample is, say, 37, then Canoco for Windows can analyse it.
6. How does the Canoco for Windows package compare with the PC-ORD 4.0 program?
PC-ORD covers some methods that are not in Canoco, for examples NMDS, TWINSPAN, and clustering analysis. Canoco is purely an ordination package with a sound theoretical basis. PC-ORD 4.0 can do some of the ordination methods available in Canoco for Windows, namely CA, DCA, PCA and, as the only canonical method, CCA. But Canoco can do more. The method that you would really miss in the PC-ORD package in practical applications is the canonical form of PCA - redundancy analysis (RDA). RDA is based on a linear model. Despite this fact, RDA often gives simpler answers to applied research problems than CCA. If you really get into modelling your data, you miss in the PC-ORD package the methods that take account of the background variation (partial ordination and partial canonical ordination). Such methods are important in impact assessment and in ecotoxicological studies, among others. Canoco has many more facilities for permutation testing than PC-ORD. A unique feature of Canoco is that its permutation tests can take account of temporal and spatial autocorrelation.
So, if you are after a mature, fully-featured implementation of ordination and constrained ordination methods with Monte-Carlo permutation tests available beyond the completely-random permutation options, we think (which is obviously biased view) that Canoco for Windows package is a better choice.
Further, to get things straight, we quote from the PC-ORD for Windows FAQ and comment where appropriate on the new Canoco version:
PC-ORD CCA vs CANOCO (v3.1)
How do CCA routines and associated Monte Carlo tests compare with those available in the most recent version of CANOCO (v3.1)?
CCA in PC-ORD is much easier to use than CANOCO, but lacks some special features. For example, CANOCO allows for inclusion of covariates while CCA in PC-ORD does not ...
We would not call inclusion of covariables features. If you, for example, need to evaluate a data set with a BACI design (often encountered in environmental impact assessment studies), you must take interaction of time and treatment as explanatory variable and use both main effects as covariables. Doing the analysis without covariables is simply incorrect. The decomposition of variance into spatial and environmental components as proposed by Borcard et al. (1992: Ecology, 73: 1045-1055) is possible only with covariables.
Also, CCA in Canoco for Windows is very easy to use.
... On the other hand, CCA in PC-ORD is integrated into a relatively comprehensive package of community analysis, providing many more classes of analyses than CANOCO. The graphics in PC-ORD are more flexible and polished than from CANOCO ...
We do not think this statement is correct any more. Please, compare the biplot example from the web pages on PC-ORD for Windows. With a similar example produced by CanoDraw for Windows 4.0.
|
PC-ORD biplot |
CanoDraw biplot
|
|
|
|
... PC-ORD has more stringent criteria for convergence of the solution than CANOCO, in response to criticism of the question of "tolerance" in CCA (see 1997 paper by Oksanen and Minchin in Journal of Vegetation Science), while this problem has not been fixed in CANOCO (as far as I know) ...
In fact, after a thorough study of the problem, the problem was solved already in CANOCO version 3.15 and the fix is available free for the registered CANOCO users. CANOCO 4.x has the fix, too. Note that the fix differs from the one suggested in the paper (see separated question in this FAQ). PC-ORD 3.0 implemented the original, partial fix and improved it in PC-ORD fix number 3.06.
... PC ORD offers two kinds of monte carlo hypotheses for CCA while CANOCO has only one. PC-ORD uses a direct reshuffling routine, similar to some versions of CANOCO (I think they may have changed that in 3.1).
Direct reshuffling is useful only for simple tests of overall hypotheses (compare overall F-tests in multiple regression). The approach used from Canoco 3.1 onwards is valid as well for partial tests. A partial test determines the significance of a particular set of environmental variables after taking into account the effects of other environmental variables. Testing a particular regression coefficient in a multiple regression equation amounts to a partial test (partial F-test or partial t-test). Testing the significance of adding an environmental variable during forward selection is another example of a partial test. The extended approach in Canoco is particularly important for environmental impact assessment studies, but also for most of the designed field experiments in ecological research.
7. I bought Canoco at Scientia Software, which is no longer distributing Canoco for Windows. Who should I ask for technical support?
Please ask you support questions at tech@microcomputerpower.com