Scientific program

Program Overview

09.00-09.10 – General introduction (Garamszegi)

Part I (chair: Garamszegi)

09.10-09.30 – Talk 1: Squeezing maximum information from matrix data in behavioural ecology (speaker: Kutsukake)

09.30-09.50 – Talk 2: The tolerance interval method for assessment of agreement in behavioral ecology studies with repeated measurements (speaker: Maurer)

09.50-10.10 – Talk 3: Repeatability for binary, proportion and count data (speaker: Nakagawa)

10.10-10.30 – Talk 4: Advances in meta-analysis in Behavioural Ecology (speaker: Santos)

10.30-11.00 – coffee break

10.50-11.10 – Talk 5: rangeMapper: A package for easy generation of biodiversity (species richness) or life-history traits maps using species geographical ranges (speaker: Dale)

Part II (chair: Nakagawa)

11.10-11.30 – Talk 6: Not quite a piece of cake: problems encountered on a behavioural ecologist’s honeymoon with Akaike’s Information Criterion (speaker: Symonds)

11.30-11.50 – Talk 7: Information-theoretic approaches to statistical analysis in behavioural ecology: an introduction to a special journal issue (speaker: Garamszegi)

11.50-12.10 – Talk 8: Avoiding common pitfalls when applying ‘animal’ models to behaviour using Bayesian methods  (speaker: Dugdale)

12.10-12.30 – Talk 9: Women have relatively larger brains than men: a comment on the misuse of GLM in the study of sexual dimorphism (speaker: Forstmeier)

12.30-12.50 – Talk 10: Using variance-covariance structures to incorporate data heterogeneity (speaker: Cleasby)

12.50-13.00 – Conclusion & take home message (Nakagawa)

13.00 – General discussion over lunch or beer

Abstracts

Overall theme: Statistical tools for Behavioral Ecologists

László Zsolt Garamszegi¹ and Shinichi Nakagawa²

¹Department of Evolutionary Ecology, Estación Biológica de Doñana-CSIC, Seville, Spain (e-mail: laszlo.garamszegi@ebd.csic.es); ²Department of Zoology, University of Otago, Dunedin, New Zealand (e-mail: shinichi.nakagawa@otago.ac.nz)

We had organized a symposium on statistical topics for the previous ISBE meeting, which subsequently generated stimulating and fruitful discussions. Since then, we have been continuing to witness that our statistical tools are developing at a high rate, and that behavioural ecologists show a huge interest in these developments and appreciate statistical dissemination. The current symposium will revolve around various issues associated with the analysis of behavioural data. First, we will visit analytical tools that have been developed for studying the repeatability and heritability of behaviour, which are fundamental for understanding the mechanisms that generate variation in behaviours within individuals as well as within species. Second, we will provide snapshots from intense discussions about the usefulness of Information Theoretic (IT) approaches especially based on AIC (Akaike’s information criterion) for behavioural ecology. Third, we will provide a useful guide to meta-analysis that has recently established itself as an essential tool for quantitative review of literature data and discuss special issues for meta-analysis in behavioral ecology. Fourth, we will investigate some important assumptions of linear models, which often remain violated resulting in misleading conclusions. However, we will keep our forum open for statistical and methodological discussions at a broader level. We aimed at collecting contributions that focus on any statistical and methodological matters that relate to behavioural ecological questions in general. Accordingly, we will also host talks on social structure analysis and geographical analysis of species.

Talk 1: Squeezing maximum information from matrix data in behavioural ecology

Nobuyuki Kutsukake

Department of Evolutionary Studies of Biosystems, The Graduate University for Advanced Studies, Hayama, Kanagawa, Japan; PRESTO, Japan Science and Technology Agency, Honcho Kawaguchi, Saitama, Japan (email: kutsu@soken.ac.jp)

Behavioural observation of social interactions or matings can be often summarized into a simple actor-receiver matrix. Here, I review a statistical toolbox for analyzing the matrix data. Association and stable group structure can be investigated by randomization methods. Mantel test and its extended version, Hemelrijk's Kr tests, solve a problem of data non-independence and have been used to test reciprocity and interchange at a group level. The Shannon-Weaver index which was originally developed in the information theory enables to quantify the degree of (un)evenness that one individual allocates behaviour to other individuals. Social network analysis provides many proxies on geometric structure of a group such as density, centrality, betweeness centrality, and so on. Squeezing maximum information from a single matrix data will help us understand complex social or mating structure in animals

Talk 2: The tolerance interval method for assessment of agreement in behavioral ecology studies with repeated measurements

Golo Maurer¹, Pankaj K. Choudhary², and Phillip Cassey¹

¹Centre for Ornithology and School of Biosciences, Birmingham University, UK (email: g.maurer@bham.ac.uk, p.cassey@bham.ac.uk); ²Department of Mathematical Sciences, University of Texas at Dallas, USA (email: pankaj@utdallas.edu) 

We describe the use of a tolerance interval method for assessing agreement between repeated measurements of continuous data. Evaluation of agreement between repeated measurements is of considerable importance in behavioral ecology. In a large number of studies, the intra-class correlation coefficient (r_I) is cited as a measure for assessing the reliability of multiple measurements on the same individual. A low value of r_I is taken to indicate low repeatability. However, it is well known that a low value of r_I may result from low variability between different individuals, not because the repeated measurements within individuals do not agree. We show that it is important to present more data than an unaccompanied value of r_I to address the repeatability of measurements. The approach of the tolerance interval method is first to model the data using a linear mixed model, and then construct the relevant asymptotic tolerance interval for the distribution of appropriately defined differences. We provide examples from two different studies: (1) laying dates in Eurasian Sparrowhawks, Accipiter nisus; and (2) claw strength in male fiddler crabs, Uca elegans. In these examples, repeated measurements were conducted for the comparison of both intra- and inter-method agreement in behavioral ecology.

Talk 3: Repeatability for binary, proportion and count data

Shinichi Nakagawa¹ and Holger Schielzeth^2,3

¹Department of Zoology, University of Otago, New Zealand (email: shinichi.nakagawa@otago.ac.nz); ²Department of Behavioural Ecology and Evolutionary Genetics, Max Planck Institute for Ornithology, Germany (email: schielz@orn.mpg.de); ³Department of Evolutionary Biology, Evolutionary Biology Centre, Uppsala University, Sweden

Repeatability (more precisely the intra-class correlation coefficient, ICC) is an important index for quantifying the accuracy of measurements and/or the constancy of phenotypes. Recently, the use of ICC became a requirement in the area of animal personality research to measure behavioural consistency. A problem of behavioural data is that they are often binary, proportion or counts (we refer to these as non-Gaussian data in relation to normally-distributed or Gaussian data). For non-Gaussian data, obtaining accurate ICCs is rather technical, unless one uses transformations of such data (although the transformation of the data leads to biased estimates of ICC). Here, we explain how we can obtain unbiased ICCs using generalized linear mixed-effects models (GLMM), which have been increasingly used by Behavioural Ecologists in recent years. We discuss a number of methods for calculating standard errors, confidence intervals and statistical significance for ICCs as well as technical difficulties arising when we calculate ICCs from non-Gaussian data. We will also introduce the R package, named rptR, which we bundled to facilitate the accurate calculations of ICCs (downloadable at https://r-forge.r-project.org/projects/rptr/).

Talk 4: Advances in meta-analysis in Behavioural Ecology

Eduardo S. A. Santos and Shinichi Nakagawa

Department of Zoology, University of Otago, New Zealand (e-mail: e.salves@gmail.com, shinichi.nakagawa@otago.ac.nz)

The use of meta-analysis by Behavioural Ecologists has been increasing since it was first used in the areas of Ecology and Evolution in the early 1990s. Despite its relatively recent appearance and common usage in Ecology and Evolution, meta-analyses have been used for over a Century in the Medical and Social sciences. Due to the head start in the usage of meta-analysis, Medical and Social scientist have been able to improve on the original method and make the meta-analytical procedure more accurate and informative. Here, we give details on some advancements that have been proposed and are in use by both Behavioural Ecologists, and Medical and Social scientists in the field of meta-analysis. We discuss 1) the use of linear mixed-effect models (LMM) to account for non-independence arising from multiple effect sizes per study as well as phylogenetic non-independence among different species, 2) meta-regression to relate the meta-analytic effects to covariates, such as study or biological characteristics, and 3) a reappraisal of the methods for detecting and correcting for publication bias in meta-analysis such as Egger’s regression and the trim-and-fill method.

Talk 5: rangeMapper: A package for easy generation of biodiversity (species richness) or life-history traits maps using species geographical ranges

James Dale¹ and Mihai Valcu²

¹Institute of Natural Sciences, Massey University, Auckland, New Zealand (e-mail: j.dale@massey.ac.nz); ²Max Planck Institute for Ornithology, Seewiesen, Germany (e-mail: valcu@orn.mpg.de)

As species numbers and geographic ranges continue to shrink or change under the influence of human consumption of natural resources, the importance of understanding geospatial patterns of biodiversity has never been greater. Traditionally however, geographical analyses of species have typically focused on understanding 1) what limits species ranges and 2) what drives variation in species diversity or richness. However another important question about geographic variation in species is what determines spatial variation in phenotypes. A classic example of this is Bergmann’s rule which states that body mass in animals tends to correlate with latitude. To facilitate the analysis of spatial variation in both biodiversity and phenotypic traits we developed a suite of R tools called rangeMapper. rangeMapper is designed for the easy generation of biodiversity (species richness) or life-history traits maps and, in general, maps of any variable associated with a species or population. The resulting raster maps are stored in a rangeMapper project file (a pre-customized SQLite database) and can thus be displayed and/or manipulated at any latter stage.

Talk 6: Not quite a piece of cake: problems encountered on a behavioural ecologist’s honeymoon with Akaike’s Information Criterion

Matthew R.E. Symonds

Department of Zoology, University of Melbourne, Victoria, Australia (e-mail: symondsm@unimelb.edu.au)

Increasingly, behavioural ecologists are applying novel model selection methods to the analysis of their data. Of these methods, information theory (IT) and in particular the use of Akaike’s Information Criterion (AIC) is becoming common. AIC allows one to compare and rank multiple competing models, and to estimate which of them best approximates the “true” process underlying the biological phenomenon under study. In theory, then, AIC provides a simple means of evaluating competing hypotheses. However, several aspects regarding the methodology and application of AIC are currently open to much debate among statisticians. These issues include the selection of candidate models and the dangers of all-subset analyses, controlling for small sample size, and the elimination of ‘uninformative’ models. I will discuss, from personal experience, how unsuspecting behavioural ecologists might stroll into the statistical line of fire, and suggest ways of dodging the bullets.

Talk 7: Information-theoretic approaches to statistical analysis in behavioural ecology: an introduction to a special journal issue

László Zsolt Garamszegi

Department of Evolutionary Ecology, Estación Biológica de Doñana-CSIC, Seville, Spain (e-mail: laszlo.garamszegi@ebd.csic.es)

Scientific thinking may require the consideration of multiple hypotheses, which often call for complex statistical models at the level of data analysis. Complex models have traditionally been treated by model selection approaches using threshold-based removal of terms, i.e. stepwise selection. A recently introduced method for model selection applies an Information Theoretic (IT) approach, which has been increasingly propagated in the field of ecology. However, a literature survey shows that its spread in behavioural ecology has been much slower, and model simplification using stepwise selection is still more widespread than IT-based model selection. Why has the use of IT method in behavioural ecology lagged behind other disciplines? A special issue (SI) will soon appear in Behavioral Ecology and Sociobiology that examines the suitability of the IT method for analyzing data with multiple predictors. The volume brings together different viewpoints to aid behavioural ecologists in understanding the method. In my talk, I will provide a brief overview on the content of the SI by emphasizing the often-misinterpreted benefits and shortcomings of the IT tool and by pointing to avenues along which the evaluation of multiple hypotheses may develop.

Talk 8: Avoiding common pitfalls when applying ‘animal’ models to behaviour using Bayesian methods

Hannah L Dugdale^1,2, David S Richardson³, Jan Komdeur¹ and Terry Burke²

¹Animal Ecology, University of Groningen, Haren, Netherlands (e-mail: h.l.dugdale@rug.nl, J.Komdeur@rug.nl); ²Department of Animal and Plant Sciences, University of Sheffield, Sheffield, UK (e-mail: h.dugdale@sheffield.ac.uk, T.A.Burke@sheffield.ac.uk); ³Centre for Ecology, Evolution and Conservation, School of Biological Sciences, University of East Anglia, Norwich, UK (e-mail: David.Richardson@uea.ac.uk)

For behaviour to evolve selection must act on behaviour; behaviour and variation in it must be heritable. In the past, behavioural ecologists have rarely tested whether behaviours are heritable, assuming instead that behaviours are flexible. This is primarily because elucidating heritability in the wild is difficult, requiring long-term study of individual behaviours and a multi-generation pedigree. As these data have become available, there has been a surge of interest in applying ‘animal’ models (mixed models that estimate how similar phenotypic traits are across related individuals) to behaviours. However, behaviours are often binary or rate measures, requiring non-Gaussian error structures and Bayesian techniques to assess the heritability of such behaviours have only recently become available. Furthermore, as relatives frequently not only share genes but also common environments, it is crucial to account for this in models. Using a long-term dataset and genetic pedigree of the Seychelles warbler, we demonstrate the application of ‘animal’ models to behaviour. We highlight how interpretation of whether helping behaviour is heritable is influenced by priors (the prior probability distribution of the unknown quantity of interest e.g. variance in helping). We therefore demonstrate the importance of using simulations to determine whether there is power to detect heritability.

Talk 9: Women have relatively larger brains than men: a comment on the misuse of GLM in the study of sexual dimorphism

Wolfgang Forstmeier

Max Planck Institute for Ornithology, Seewiesen, Germany (e-mail: forstmeier@orn.mpg.de)

General linear models (GLM) have become such universal tools of statistical inference, that their applicability to a particular data set is rarely questioned. These models are designed to minimize residuals along the y-axis, while assuming that the predictor (x-axis) is free of statistical noise (ordinary least square regression, OLS). However, in practice, this assumption is often violated, which can lead to erroneous conclusions, particularly when two predictors are correlated with each other (e.g. sex and body size in size dimorphic species). This is best illustrated by two examples from the study of allometry, which have received great interest: (1) the question of whether men or women have relatively larger brains after accounting for body size differences, and (2) whether men indeed have shorter index fingers relative to ring fingers (digit ratio) than women. These examples clearly illustrate that GLMs produce spurious sexual dimorphism in body shape where there is none (e.g. relative brain size). Likewise, they may fail to detect existing sexual dimorphisms in which the larger sex has the lower trait values (e.g. digit ratio) and, conversely, tend to exaggerate sexual dimorphism in which the larger sex has the relatively larger trait value (e.g. most sexually selected traits). These artifacts can be avoided with reduced major axis regression (RMA), which simultaneously minimizes residuals along both the x and the y-axis.

Talk 10: Using variance-covariance structures to incorporate data heterogeneity

Ian R Cleasby

Department of Animal & Plant Sciences, University of Sheffield, Sheffield, UK (e-mail: bop06irc@sheffield.ac.uk)

Whenever a researcher applies a linear regression to their data they implicitly make a host of assumptions. These assumptions must be verified to ensure that we can trust the results obtained. One particular problem that can arise when conducting a linear regression is the occurrence of heterogeneity. Heterogeneity occurs when the variance of the residuals is not constant. Although heterogeneity does not cause coefficient estimates to be biased it does affect the standard error of these estimates. The most common method for dealing with heterogeneity is to transform the data. However, in many cases transforming data may not be desirable as it makes results harder to interpret and because heterogeneity may provide important ecological information. Here, I briefly discuss how we can use variance-covariance structures to successfully incorporate heterogeneity within statistical models as an alternative to data transformation. The key idea is that we can include information on the spread of residuals within a model. To date, explicit specifications of variance-covariance structures do not appear to be widely used in behavioural ecology, which suggests that such a technique may not be widely appreciated. However, there are a number of situations when the use of this technique may prove useful and they should be considered by researchers as a means of dealing with heterogeneity. I also discuss extensions of the use of variance-covariance structures in regression methods in wider contexts – such as dealing with spatial, temporal and phylogenetic covariance, which are often encountered in ecological data.