Should we include a threshold option when assessing the strength or the significance of interactions?

[jmalcan]

When estimating interaction strength through indices (associndex), we have implemented the possibility of using a threshold value of recruitment density, in order to remove possible outlayer cases that can be the result of sampling limitations. For example, an outstandingly large estimate of density can occur if a canopy species that is very scarce in the sampled plots (e.g. one that is represented by a single individual) happens to have some recruits beneath its canopy just by chance (e.g. the individual is very close to a large fruiting plant or another species). Such high densities are not a property of the canopy species or of the canopy-recruit interaction, so they should be discarded when estimating the strength of interactions. A good sampling design can minimise this problem, but it cannot be totally prevented.
Ultimately, the problem is how to distiguish this kind of outlayers from truly outstanding nurse plants or strong interactions, if it is not possible to get more data or conduct experiments. In some particular cases, one may find some ecological argument to support the credibility of the estimates (e.g. evidence from other studies indicating also that the species is a good nurse). However, in most cases it will not be possible. Thus, I propose a conservative approach and use a statistically determined threshold to avoid overestimating interaction strength.
For the same reason, I propose including a threshold option in functions testing pairwise interactions (int_significance) and node-level interactions (canopy_service and recruitment_niche tests), since these tests are based on recruitment densities (i.e. recruitment frequencies relative to canopy cover).