Science is under attack in many parts of the world today, but I would like to call attention to an internal enemy of scientific advancement that is equally worrisome. I will address my comments to ecological papers that employ correlations. I am sure that the errors I discuss will not ever be found in other sciences (see Ioannidis 2019).
Perhaps the most common statistic to be found in ecological papers is the correlation coefficient r. The correlation coefficient measures the strength of the association between two variables. Typically, in ecological use the two variables ought to have some mechanistic or causal relationship between them. This is the key question you must ask before calculating an r value. Many examples of nonsense correlations are given in Huff (1954) and Best (2001). A 20-year data set of the number of people attending a particular church in a specified area may be very highly correlated with the increase in automobile accidents in this area. We laugh at these sorts of examples but the central problem in ecology is the measure of cause and effect and how to decipher the effects of multiple causes on our variable of interest. Tree growth may be correlated with the amount of summer rainfall, but it may also be correlated with summer temperatures. How we proceed from here is determined by the purpose of the investigation. If you wish trees to grow better in a drought, be sure to water them. It may not matter to you what the temperature is in this study, particularly if you have no control of temperature but can control water. But if the purpose of the investigation is to determine all the factors that control tree growth, you will have a much more complex problem to resolve, concerning the chemistry of the soil and the genetic makeup of each tree, among other variables. In all these cases the key is to specify what is the question you wish to answer?
Much of ecological research is exploratory, step 1 of trying to understand species, populations or communities. Trap 1 which you must try to avoid is measure everything possible. Falling into trap 1 leads to trap 2 which is you have much data with little insight. It is certainly possible to be lucky and gain a critical insight by measuring many variables. My impression which may be incorrect is that frequently you have missed some critical interaction. Everyone I know has made this mistake, but you can recover from it by becoming more specific in your aims.
At this point we can, if possible, move into experimental ecology in which we vary treatments to some populations or communities and have controls that are not manipulated. One outline of experimental approaches in wildlife management is given in Hone and Krebs (2023). But the warning is clear even with supposed experimental treatments of modern medicine where the consequences of mistakes come very close to home (Ioannidis 2021).
The take-home message from all these warnings is that when you discuss correlations be concerned with the variables that are omitted from your analysis, and because of that be cautious about your conclusions. The best way to indicate this in your writing is to present in your conclusions a simple statement of what needs to be done next on this ecological problem.
Best, J. 2001. Damned Lies and Statistics: Untangling Numbers from the Media, Politicians, and Activists. University of California Press. 204 pp.
Hone, J. and Krebs, C.J. 2023. Causality and wildlife management. Journal of Wildlife Management 2023: e22412. doi:10.1002/jwmg.22412.
Huff, D. 1954. How to Lie With Statistics. R.S. Means Company pp. 142.
Ioannidis, J.P.A. 2019. What have we (not) learnt from millions of scientific papers with P values? American Statistician 73(sup1): 20-25. doi:10.1080/00031305.2018.1447512.
Ioannidis, J.P.A. 2021. Hundreds of thousands of zombie randomised trials circulate among us. Anaesthesia 76(4): 444-447. doi:10.1111/anae.15297.