Griffen et al. (2016) proposed that studies should be designed:

• to lead to a mechanistic understanding of the observed responses (rather than just description);
• to  parameterize and inform population and ecosystem models, in order to allow upscaling from the experiments to field populations, communities and ecosystems;
• with a sufficient number of levels of the driver(s) to inform about the presence of linear or non-linear responses including thresholds to the drivers;
• conclusions about effect sizes and interactions (additive or multiplicative) need to consider the data model used, as data transformation changes the statistical model being tested.

Useful references

Griffen BD, Belgrad BA, Cannizzo ZJ, Knotts ER, Hancock ER (2016) Rethinking our approach to multiple stressor studies in marine environments. Marine Ecology Progress Series 543:273-281.

Munns WR Jr (2006) Assessing risks to wildlife populations from multiple stressors: overview of the problem and research needs. Ecol Soc 11: 23