AUTHOR=Yang Jing , Lu Jiahui , Chen Yue , Yan Enrong , Hu Junhua , Wang Xihua , Shen Guochun 

TITLE=Large Underestimation of Intraspecific Trait Variation and Its Improvements

JOURNAL=Frontiers in Plant Science

VOLUME=Volume 11 - 2020

YEAR=2020

URL=https://www.frontiersin.org/journals/plant-science/articles/10.3389/fpls.2020.00053

DOI=10.3389/fpls.2020.00053

ISSN=1664-462X

ABSTRACT=Intraspecific trait variation (ITV) widely exists in natural communities and has gained increasing attention due to its large ecological effects on community dynamics and ecosystem functioning. However, the estimation of ITV per se has yet to be received much attention, despite an accurate estimation of ITV is essential to make reasonable trait-based ecological inferences. It remains unclear whether and to what extent our current estimation of ITV would biased. Therefore, we focused on how to accurately quantify ITV via the coefficient of variation (CV), because literature survey showed CV is the most commonly used method in directly quantifying ITV and is dimensionless that can be compared among traits, species and studies. We asked which is the best CV estimator and what are the minimum sample size that could reach ±5% and ±10% accuracies of the ITV estimate for a completely random sample scheme. To answer these questions, we proposed four new composite estimators and compared their performance with four existing CV estimators across various sample sizes using both a simulated and three large real trait datasets from local to regional scales. Our results consistently showed estimates of ITV by the most commonly used estimator, 〖CV〗_1=σ_sample/μ_sample, is often underestimated, and this underestimation varies largely among traits/species and can reach 49.8%. This bias arises from an oft-ignored fact that CV1 is a biased estimator and the extent of bias depends on the sample size, skewness and kurtosis of the trait value distribution. Our new composite estimator CV5 takes into consideration of these dependencies and without any specific assumption of the trait value distribution. It can largely reduce the underestimation of ITV and reached ±5% and ±10% accuracies once the sample size were greater than 70 and 45 in most of situations, respectively. These results demonstrated that many of our previous ITV estimates could be largely underestimated and these underestimations are not equal among species and traits even using the same sample size. Our composite estimators are simple but effective in the improvement of the ITV estimation, thereby facilitating a more accuracy understanding of ITV in community structures and dynamics.