2 Management and Analysis of Wildlife Biology Data bret a. collier and thomas w. schwertner INTRODUCTION HE GENERAL FOCUS in this chapter is on outlining the range of op- tions available for the management and analysis of data collected during wildlife biology studies. Topics presented are inherently tied to data analy- Tsis, including study design, data collection, storage, and management, as well as graphical and tabular displays best suited for data summary, interpretation, and analysis. The target audience is upper-level undergraduate or masters-level gradu- ate students studying wildlife ecology and management. Statistical theory will only be presented in situations where it would be beneficial for the reader to understand the motivation behind a particular technique (e.g., maximum likelihood estima- tion) versus attempting to provide in-depth detail regarding all potential assump- tions, methods of interpretation, or ways of violating assumptions. In addition, po- tential approaches that are currently available and are rapidly advancing (e.g., applications of mixed models or Bayesian modeling using Markov Chain Monte Carlo methods) are presented. Also provided are appropriate references and pro- grams for the interested reader to explore further these ideas. The previous authors (Bart and Notz 2005) of this chapter suggested that before collecting data, visit with a professional statistician. Use the consultation with a professional to serve as a time to discuss the objectives of the research study, out- line a general study design, discuss the proposed methods to be implemented, and identify potential problems with initial thoughts on sampling design, as well as to determine what statistical applications to use and how any issues related to sam- pling will translate to issues with statistical analysis. Remember, no amount of sta- tistical exorcism can save data collected under a poor study design. NATURE OF KNOWLEDGE The primary goal of science is to accumulate knowledge on what reliable state- ments regarding the state of nature can be made. However, defining “knowledge” and “reliable” with respect to science has proven to be a difficult task (Kuhn 1996). Epistemology is the branch of philosophy that focuses on determining what con- stitutes knowledge and how knowledge is acquired. It draws contrasts between the 2 primary methods of knowledge acquisition—rationalism and empiricism. Ratio- nalists argue that reason alone is sufficient for knowledge accrual (Morrison et al. 2008:4). Conversely, knowledge acquired through experience underlies the empiri- bret a. collier and thomas w. schwertner cist perspective of knowledge acquisition. Scientific research sign used, certain elements of the target population have in general and wildlife ecology in particular tends to rely probability zero of being included in the sampled popu- most heavily on empiricism. lation; Williams et al. 2001). In this case it may be possible Judgment of reliability is another matter. Most wildlife to develop a predictive model that seems to provide an accu- ecologists use the process of induction (generalizing from rate representation of population processes. However, the specific results; Guthery 2008) or take the results from a sin- model is unreliable, as it only predicts to a subset of the gle or group of studies and apply those results widely. Op- overall (sampled) population. If inferences are to be made posed to induction is deduction, or generalizing a principle to the target population in this case, then auxiliary data to occurrence of a specific event or events. A good example must be collected that shows a consistent relationship be- of deduction might be that all animals are mortal, deer are tween the sampled and target populations (for further dis- animals, and thus deer are mortal (Morrison et al. 2008:11). cussion see Williams et al. 2002a, Morrison et al. 2008). Retroduction occurs when deductions are used to deter- The framework for all analysis approaches discussed in this mine what conditions might have lead to the observed re- chapter hinge on appropriate sampling designs. Therefore, sult (Guthery 2008). Retroduction is common in discussion a professional statistician should be consulted before sections of publications, where scientists try to determine study implementation. Additionally, readers are referred what mechanism lead to the observed results. For example, to Chapter 1, This Volume) for a discussion of sampling in Dreibelbis et al. (2008) used retroduction from a set of ob- ecological studies and to the literature cited in this chapter servations (missing egg shell fragments, no evidence of scat for books and articles addressing sampling designs and sta- or hair) to deduce that a snake was the most likely predator tistical inference. of turkey nests under this set of observations. In the inductive framework, analyzing data serves to gen- DATA COLLECTION eralize data collected from the samples for use as surrogates for describing the population as a whole (Romesburg 1981, What to Collect Guthery 2008). Inference is made because there is some bio- To begin a wildlife ecology study, first consider what data to logically, socially, or politically motivated interest in what collect to make the study as fruitful as possible. It is impera- characterizes a population and what makes it different from tive to “look before you leap” when deciding on the relevant other populations, and because knowledge of this difference data. Time spent collecting data that have limited impact on would be beneficial to management. the question of interest should be avoided (e.g., ground lit- ter type when you are estimating canopy bird abundance). REQUIREMENT FOR STATISTICAL INFERENCE Also, data should not be collected that is of different resolu- tion than the question of interest (e.g., hourly precipitation One of the first issues all ecologists are faced with during information when interested in estimates of annual elk [Cer- the initial stage of study development is determining how vus elaphus] survival). Data collection is a delicate balancing data of interest will be collected. In the rush to start field act requiring care to include potentially useful data for ex- work, adequate time and resources may not be assigned to plaining the biological process being studied while omitting evaluating potential analytical approaches and combining irrelevant information. Collect data with future analysis in those approaches with sampling designs to address ques- mind, remembering current time and money limitations. tions of interest. However, appropriate sampling procedures are paramount, because researchers cannot make direct ob- How to Collect servations of every unique individual in the population (e.g., Without a doubt one of the most important yet unappreci- a census). In practice, the objective of sampling is to gener- ated aspects of wildlife research is the choice of collection alize population processes based on data collected from a instrument. Rarely does data sheet development get dis- subset of the population of interest (Cochran 1977, Thomp- cussed in data analysis chapters. Yet data collection proto- son 2002a, Morrison et al. 2008). Although it is nearly always cols and instruments should be given as much consideration impossible to census populations, by taking a sample ecolo- as other parts of study development, as they are a critical gists can make inferences about the population in question. link in a chain connecting the study subjects with the study The importance of appropriate sampling cannot be over- results, the failure of which could compromise the reliabil- emphasized when considering data analysis. Predictive mod- ity of any knowledge drawn from the research. els, and hence predictions, will be more accurate and useful There are several considerations that students and scien- when data collection is based on an appropriate sampling tists need to take into account when developing data proto- design (Harrell 2001). The relationship between analysis and cols. As an example, consider a 2-year masters project, dur- sampling design is clear: population-level inferences require ing which data collection will occur over several seasons and data that represent the population of interest. Consider the the study species will undergo several phenological changes. situation wherein the sampled population is not equivalent When developing data protocols, determine whether a sin- to the target population (e.g., because of the sampling de- gle data sheet for species-specific data collection across all MANAGEMENT AND A N A LY S I S OF WILDLIFE BIOLOGY DATA 43 seasons (e.g., breeding or wintering) will be used, or are wherein there is interest in evaluating how prescribed fire separate sheets needed for each season? Will 1 data sheet influences avian survival over the course of an 8-week per individual per day be used, or will individuals (or sample breeding season. Initially, determine how frequently surveys units) be lumped by day onto a single data sheet? Will the will be conducted (perhaps through radiotracking). If inter- same data sheets for each period under study be used? Will ested only in basic survival, then locate all the radiotagged data collection occur during a single season over several individuals at the end of the eighth week, and the number years, or will data collection be continuous over several sea- still alive divided by the number radiotagged at the begin- sons, requiring different data types to be collected?