These variables are apparently all independent variables in a strict sense.

Aggressiveness is a subject property you do not impose to your P. infestans, unless you are doing gene-engineering (see below if it is the case). You probably select some which are more aggressive than others to form conditions and test these aggressiveness levels under various (or under similar=constant) conditions; but the variable aggressiveness remains selected here and thus quasi-experimental. You don’t give much details on your research, but it is apparently plainly correlational (it is not an experiment proper).

If you impose aggressiveness by gene-engineering, then you create strains showing various levels of aggressiveness. Aggressiveness in this case perfectly satisfies the definition of a genuine independent variable and it can serve in a true experiment with various levels (low aggressiveness, high, ... as defined by the SNPs you have engineered) . Measures taken of their devastating effects on the plants would be the dependant variables (these effects depend of the independent variable, they are presumably caused by the independant variable)

Still, even if you select cases, under a regressive schema (e.g. multivariate regressive analysis and the like) you can call by convention aggressiveness as the independent variable (with various levels varied through selection of appropriate individuals) and the measure of damage these fongi would cause on the plants as dependent variables. You could thus regress aggressiveness onto measures of damage. But aggressiveness of the P. infestans has to be qualified or measured INDEPENDANTY and usually BEFORE introducing them to their host where you want to measure their devastating effect.

Dear Dr. Beaugrand,

can it be inferred from your detailed explanation that aggressiveness components such IF, LP, LA, LGR and RAULEC I should consider as dependent variables?

In a multivariate analysis, the choice of the term independent vs dependent variable is conventional and/or guided by the model of (presumed) causality you want to test trough regression/correlation. The variables whose effect is presumed to exist on other variables are momentarily named Independent variables and the ones presumably affected by the IVs are called dependent variables (DVs).

But some DVs could well serve as IVs when testing another model.

From what you write the outcome of your model is the "damage" to potato leaves caused by different strains of P. infestans.

In this case infection efficiency, lesion area, lesion growth rate and area under disease are different manifestations of that damage so they are each a dependent variable. I.e. a slightly different variation of your model.

You could/should test if you can simplify your "damage" model by combining these variables using factor analysis.

The independent variables are the underlying characteristics of P. infestans you want to test (e.g. gene differences, environmental factors, etc.) for which you should also use factor analysis to see if they are truly independent of each other.

Thanks Nicolas,