Mikaela,
Good for you to ask such a question! Most individuals don't even think
twice before extrapolating. The most notable examples that comes to mind are
the majority of climate change studies being done at present. Predict far
enough into the future and you are doubtlessly extrapolating outside the bounds
of your data. Does that mean that we shouldn't do it? I would contend that
extrapolation is acceptable under certain circumstances:
1) A prediction is better than none at all. Decisions will be made with or
without our scientific predictions. I would prefer that they be based on the
best available science.
2) It is possible to extrapolate judiciously. Do your data follow a linear
trend? Is it ecologically reasonable for that trend to continue beyond the
bounds of your data? Consider a simplified example: Is it ok to assume that a
plant species response to temperature will follow a similar trajectory if that
temperature goes from 18 to 20 degrees C? Probably. On the other hand, what
if the temperature goes from -1 to 1 degrees C? Here an important threshold
(the freezing point of water) has been crossed, and we need to use our
biological, not statistical training to guide us.
Having said that, if you do extrapolate beyond the range of your data, you are
introducing additional uncertainty into your results. Recognize and address
this uncertainty in your interpretation of the results.
Good luck,
Candan Soykan
-----Original Message-----
From: Ecological Society of America: grants, jobs, news
[mailto:[email protected]] On Behalf Of Mikaela Gioia Howie
Sent: Thursday, March 06, 2014 6:25 AM
To: [email protected]
Subject: [ECOLOG-L] Is it ever appropriate to extrapolate past your data?
Hi Eco-loggers,
I was just wondering if anyone would like to weigh in on the issue of
extrapolation past one's data in order to make a prediction? Can this be
appropriate for a peer-reviewed publication? What other methods can be
employed to make simple predictions rather than extrapolation? Basically, I am
using regression models to predict what would happen at a point that is outside
of my dataset. I realize that this is would not be a robust prediction but I
have not determined another method to make a similar prediction.
Thanks for your help!
Mikaela
Mikaela G. Howie
