Trace:

en:confusions

This shows you the differences between two versions of the page.

Both sides previous revision Previous revision | |||

en:confusions [2019/11/24 17:32] David Zelený [Common confusions and mistakes] |
en:confusions [2019/11/24 17:37] (current) David Zelený [Use “envfit” to project explanatory variables into RDA or CCA and to test their significance] |
||
---|---|---|---|

Line 32: | Line 32: | ||

Longer explanation follows below, with examples. | Longer explanation follows below, with examples. | ||

- | First, to clarify. One of the use of the function ‘’envfit’’ (library vegan) is to calculate regression of supplementary variables to ordination axes in unconstrained ordination, in order to help with interpretation of these axes. Say we have community data and three external environmental variables; we calculate unconstrained ordination using the community data, and then we aim to interpret the ecological meaning of unconstrained ordination axes with available environmental variables by calculating their regression. These variables can then be projected onto the ordination diagram as supplementary (not explanatory!), and regression of each variable independently can be tested by permutation test (this is what ‘’envfit’’ returns). But variables which are significant here are not those which are important for species composition, but those which have the best fit to variation extracted by unconstrained ordination into the main (usually first and second) ordination axes. Often they are important, but not necessarily. | + | First, to clarify. One of the use of the function ‘’envfit’’ (library vegan) is to calculate regression of supplementary variables to ordination axes in unconstrained ordination, in order to help with interpretation of these axes. Say we have community data and three external environmental variables; we calculate unconstrained ordination using the community data, and then we aim to interpret the ecological meaning of unconstrained ordination axes with available environmental variables by calculating their regression. These variables can then be projected onto the ordination diagram as supplementary (not explanatory!), and regression of each variable independently can be tested by permutation test (this is what ‘’envfit’’ returns). But variables which are significant here are not those which are important for species composition, but those which have the best fit to variation extracted by unconstrained ordination into the main (usually the first and the second) ordination axes. Often these variables are important, but not necessarily. |

- | Now, if you calculate constrained ordination, then environmental variables enter the analysis, and resulting (constrained) axes already contains information about them. If now you use ‘’envfit’’ to calculate regression of these same variables to constrained axes, you are likely to get highly significant results; you are regressing two variables (environmental variable vs samples scores on ordination axes) which contain the same information (since axes were calculated also using the environmental variable). | + | Now, if you calculate constrained ordination, then environmental variables enter the analysis, and resulting (constrained) axes already contain information about them. If now you use ‘’envfit’’ to calculate regression of these same variables to constrained axes, you are likely to get highly significant results; you are regressing two variables (environmental variable vs samples scores on ordination axes) which contain the same information (since axes were calculated also using these environmental variables). |

en/confusions.txt · Last modified: 2019/11/24 17:37 by David Zelený