Animal personality: what are behavioural ecologists measuring?

doi:10.1111/BRV.12007

Journal Article•DOI•

Animal personality: what are behavioural ecologists measuring?

Alecia J. Carter¹, Alecia J. Carter², William E. Feeney¹, Harry H. Marshall², Harry H. Marshall³, Guy Cowlishaw², Robert Heinsohn¹ - Show less +3 more•Institutions (3)

Australian National University¹, Zoological Society of London², Imperial College London³

01 May 2013-Biological Reviews (Wiley)-Vol. 88, Iss: 2, pp 465-475

TL;DR: An integrative theoretical framework is suggested that incorporates the tools that were developed to overcome similar methodological problems in psychology to facilitate a robust and unified approach in the study of animal personality.

read less

Abstract: The discovery that an individual may be constrained, and even behave sub-optimally, because of its personality type has fundamental implications for understanding individual- to group-level processes. Despite recent interest in the study of animal personalities within behavioural ecology, the field is fraught with conceptual and methodological difficulties inherent in any young discipline. We review the current agreement of definitions and methods used in personality studies across taxa and systems, and find that current methods risk misclassifying traits. Fortunately, these problems have been faced before by other similar fields during their infancy, affording important opportunities to learn from past mistakes. We review the tools that were developed to overcome similar methodological problems in psychology. These tools emphasise the importance of attempting to measure animal personality traits using multiple tests and the care that needs to be taken when interpreting correlations between personality traits or their tests. Accordingly, we suggest an integrative theoretical framework that incorporates these tools to facilitate a robust and unified approach in the study of animal personality.

...read moreread less

Summary (2 min read)

Jump to: [2.1 Random effect formulation of the LASSO] – [3.2.1 Bone density data] – [3.2.2 Breast cancer data] and [4 Discussion]

2.1 Random effect formulation of the LASSO

The maximum a posteriori (MAP) estimator under the random effects model in (4), which is conditional on σ 2 , is equivalent to the estimator minimizing the LASSO objective function with external information driven penalization in (3).
The population variance parameter σ 2 can estimated from the data or given a point-mass prior (Li and Lin, 2010) .
Park and Casella (2008) and Li and Lin (2010) used a specification that assigns a non-informative prior for σ 2 and used a Gibbs sampler based on full conditional distributions to obtain regression coefficients estimates.
The authors method is different in that, rather than extending the model to include Bayesian inference over the hyperparameters, the authors use a empirical Bayes approach that maximize the log likelihood of hyperparameters.
That is, the hyperparameters α are estimated from the data by first marginalizing over the coefficients β and then performing what is commonly referred to as empirical Bayes, evidence maximization or type-II maximum likelihood Tipping (2001).

3.2.1 Bone density data

The data were randomly split into a training data consisting of 80% of the observations and a test data set consisting of 20% of the observations.
The authors fitted the adaptive LASSO, standard LASSO and their proposed method in the training data and evaluated their prediction performance in the testing data.
The authors repeated 100 random splits of the full data into training and test sets.
Figure 2 shows the MSE, R 2 and the number of selected (non-zero) expression features across the 100 splits.
To gain further insight into the prediction performance results, the authors examined the penalties applied to the regression coefficients by each of the methods when fitted on the full data.

3.2.2 Breast cancer data

The authors compared the xtune LASSO incorporating the meta-features described above, with the standard and the adaptive LASSO.
As in the first example, standard LASSO was tuned by repeated 10 fold cross-validation and the adaptive LASSO is implemented using the adalasso function in the parcor R package.
Table 1 compares the AUC, the number of selected features and the computa-13 tion time for the standard, the adaptive and the xtune LASSO.
In agreement with their simulation results, the xtune LASSO also yielded a much more parsimonious model with only 10 selected features while the standard LASSO selected 207 features.

4 Discussion

A related limitation is that in its current implementation xtune does not scale to ultra high dimensional datasets.
Typical datasets that xtune LASSO can currently handle have sample size of up to n 5000, with p 50, 000 features and q 100 meta-features.
To further widen the range of applicability of xtune, the authors are pursuing extensions to binary (logistic regression) and time to event (Cox regression) outcomes, as well as the incorporation of the Ridge and Elastic-Net penalties in addition to the LASSO.

Did you find this useful? Give us your feedback

Content maybe subject to copyright Report