Animal personality: what are behavioural ecologists measuring?
Summary (2 min read)
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
Citations
245 citations
207 citations
Cites background from "Animal personality: what are behavi..."
...Personality variation refers to individual differences in behaviour, such as risk-taking or aggression, which are consistent over time and across contexts (Wilson et al., 1994; Gosling, 2001; Carter et al., 2013), and are undoubtedly ubiquitous (Sih et al....
[...]
...Personality variation refers to individual differences in behaviour, such as risk-taking or aggression, which are consistent over time and across contexts (Wilson et al., 1994; Gosling, 2001; Carter et al., 2013), and are undoubtedly ubiquitous (Sih et al., 2004)....
[...]
198 citations
188 citations
Cites background from "Animal personality: what are behavi..."
...In order to assess a trait (which is stable over time), the test(s) used must yield highly replicable results (Carter et al., 2013)....
[...]
...A trait thus is considered a permanent characteristic, whereas a state is considered as a transient condition that is only observable at particular moments (see also Fridhandler, 1986; Koski, 2011; Carter et al., 2013)....
[...]
...Repeatedly testing emotional state (e.g., across the lifespan of an animal) may yield information about its emotional trait, i.e., the behavior indicative for a particular trait needs to be repeatable (Carter et al., 2013)....
[...]
187 citations
Cites background from "Animal personality: what are behavi..."
...The few studies that have investigated correlations among different personality measures that were thought to quantify the same trait have found some surprising results (Carter et al. 2013; Garamszegi et al. 2013)....
[...]
References
15,795 citations
"Animal personality: what are behavi..." refers background or methods in this paper
...While each approach has benefits and drawbacks (Koski, 2011), the high correspondence between measures suggests simultaneous application of these methods may be beneficial (Uher & Asendorpf, 2008), at the very least to test convergent validity (Campbell & Fiske, 1959)....
[...]
...Finally, there is much historical (and perhaps contemporary) conceptual debate within the psychometric literature (see definitions in Cronbach & Meehl, 1955; Campbell & Fiske, 1959 and discussion in Uher, 2011a)....
[...]
...…psychology (McCrae, 1982; Duckworth & Kern, 2011) therefore emphasise the importance of attempting to measure animal personality traits using multiple methods, or in the case of behavioural ecology, multiple tests (Cronbach & Meehl, 1955; Campbell & Fiske, 1959; Uher, 2011a; Weiss & Adams, 2013)....
[...]
...Biological Reviews © 2012 Cambridge Philosophical Society the measurements (Campbell & Fiske, 1959)....
[...]
...However, we could alternatively conclude that the two tests measured different personality traits (Burns, 2008; Carter et al., 2012b), and so are highly discriminant (Campbell & Fiske, 1959; for alternative interpretations, see Cronbach & Meehl, 1955)....
[...]
9,935 citations
"Animal personality: what are behavi..." refers background or methods in this paper
...Finally, there is much historical (and perhaps contemporary) conceptual debate within the psychometric literature (see definitions in Cronbach & Meehl, 1955; Campbell & Fiske, 1959 and discussion in Uher, 2011a)....
[...]
...…psychology (McCrae, 1982; Duckworth & Kern, 2011) therefore emphasise the importance of attempting to measure animal personality traits using multiple methods, or in the case of behavioural ecology, multiple tests (Cronbach & Meehl, 1955; Campbell & Fiske, 1959; Uher, 2011a; Weiss & Adams, 2013)....
[...]
...However, we could alternatively conclude that the two tests measured different personality traits (Burns, 2008; Carter et al., 2012b), and so are highly discriminant (Campbell & Fiske, 1959; for alternative interpretations, see Cronbach & Meehl, 1955)....
[...]
...…personality research is to maintain its rate of progress, it must develop a robust methodology including multiple trait tests, reliability and validation (Cronbach & Meehl, 1955; Uher, 2011a; Weiss & Adams, 2013), and continue to identify research questions and hypotheses clearly at their outset....
[...]
...That is, a construct is an a priori theoretical idea of how behaviours should be inter-correlated (Cronbach & Meehl, 1955; Nunnally, 1978)....
[...]
5,061 citations
4,777 citations
"Animal personality: what are behavi..." refers methods in this paper
...This concept, primarily employed by differential and human personality researchers, uses factor analysis or principal component analysis (PCA) to identify the orthogonal personality factors that are robust across investigations, samples and time (Goldberg, 1992, 1993)....
[...]
4,025 citations
"Animal personality: what are behavi..." refers methods in this paper
...This concept, primarily employed by differential and human personality researchers, uses factor analysis or principal component analysis (PCA) to identify the orthogonal personality factors that are robust across investigations, samples and time (Goldberg, 1992, 1993)....
[...]
...This allows explanation of each factor’s observed patterns of variation and can be used to predict various aspects of an individual’s behaviour such as job-related performance (Goldberg, 1993)....
[...]