The final paper from my PhD has been published!
Our intent for this paper was to encourage thorough testing of multiple growth model forms and an increased emphasis on assessing model fit relative to a model’s purpose. Our paper is open access and accompanied by open access code (thanks Jian Yen) to do everything we wrote about!
During my PhD I was interested in seeing if we could use plant functional trait data to predict plant growth through time across multiple species.
Some battles involved with this quest were determining which growth models to use, which predictor variables to use and how to evaluate all of this relative to our objectives.
It was actually relatively easy to find a growth model that described any one plant species’ growth pretty well. However, I didn’t want to find the best growth model for each individual species – I wanted to find one growth model to describe all my species’ growth. I also actually didn’t really want to just describe my plant species’ growth – I really wanted to predict the growth of new species.
To do this, I had to work out which models could describe the growth of one species adequately, describe the growth of all species adequately, predict the growth of one species adequately and predict the growth of all species adequately both inside and outside of my dataset.
I also had to figure out what I meant by ‘adequately’.
In this paper, we outline the methods we used to model and evaluate predictive trait-based models of growth for multiple plant species.
We use three data sets on plant height over time and two validation methods—in-sample model fit and leave-one-species-out cross validation—to evaluate nonlinear growth model predictive performance based on functional traits.
In-sample measures of model fit differed substantially from out-of-sample model predictive performance; the best fitting models were rarely the best predictive models. Careful selection of predictor variables reduced the bias in parameter estimates and there was no single best model across our three data sets. Testing and comparing multiple model forms is important.
Again, our intent is to encourage thorough testing of multiple growth model forms and an increased emphasis on assessing model fit relative to a model’s purpose.
We hope to contribute to the practice of growth modeling by developing methods and code for the evaluation of predictive capacity of non-linear growth models.
This paper is accompanied by an R package, growmodr, to fit and validate nonlinear growth models (available at <https://github.com/jdyen/growmodr ). An example of fitting and validating a growth curve model is in Appendix 1.
Thanks to Jian Yen and Pete Vesk for accompanying me on this paper’s journey .. it was mostly fun!
Here are some of the plants we were working with: