Backward Stepwise Linear Regression

Backward stepwise regression starts with all explanatory variables included the model. It then removes the least significant explanatory variable, that is, the one with the highest p-value, at each step, until all variables have been added. By scrutinising the overall fit of the model variables will be automatically removed until the optimum model is found.

Load both your species and your environmental datasets as described in Opening a Data Set

From the Regression menu on the Toolbar select Backward Stepwise. A list of dependent variables (usually species) is displayed.

Select the species against which you wish to test your environmental variables.

Click OK to run the analysis and display the results on the Regression tab.

The results report shows the sequences of the procedure as Steps:

Step 1 - Shows the effect of including all the explanatory variables into the model, with the individual p-values.

Step 2 - Shows the effect of removing the least explanatory variable from the model.

...and so on.

In most species/environmental datasets it is unusual for more than three variables to be included in any explanatory model. 

A common problem with environmental data sets is that of multicollinearity between explanatory variables. ECOM automatically checks for multicollinearity and, if detected, will take to you a screen that details the problem and allows you to remove one of the correlated variables.