USGS - science for a changing world

Northern Prairie Wildlife Research Center

  Home About NPWRC Our Science Staff Employment Contacts Common Questions About the Site

Symposium:
Proceedings of the North Dakota Academy of Science

Carnivore Scent-station Surveys: Statistical Considerations

Glen A. Sargeant* and Douglas H. Johnson

Department of Wildlife Ecology, University of Wisconsin,
226 Russell Labs, 1630 Linden Drive, Madison, WI 53706 (GAS)
and United States Geological Survey, Biological Resources Division,
Northern Prairie Science Center, 8711 37th St. SE, Jamestown, ND (DHJ)


This resource is based on the following source (Northern Prairie Publication 1015):
Sargeant, Glen A., and Douglas H. Johnson.  1997.  Carnivore scent-station surveys:
     statistical considerations.  Proceedings of the North Dakota Academy of 
     Science 51:102-104.
This resource should be cited as:
Sargeant, Glen A., and Douglas H. Johnson.  1997.  Carnivore scent-station surveys: 
     statistical considerations.  Proceedings of the North Dakota Academy of 
     Science 51:102-104.  Jamestown, ND: Northern Prairie Wildlife Research Center 
     Online.  http://www.npwrc.usgs.gov/resource/mammals/carnivor/index.htm
     (Version 17DEC97).

Scent-station surveys are a popular method of monitoring temporal and geographic trends in carnivore populations. We used customary methods to analyze field data collected in Minnesota during 1986-93 and obtained unsatisfactory results. Statistical models fit poorly, individual carnivores had undue influence on summary statistics, and comparisons were confounded by factors other than abundance. We conclude that statistical properties of scent-station data are poorly understood. This fact has repercussions for carnivore research and management. In this paper, we identify especially important aspects of the design, analysis, and interpretation of scent-station surveys.

Introduction

Animal abundance is one appropriate measure for gauging the success of wildlife management, monitoring the status of threatened and endangered species, and determining the outcome of many experiments. Thus, estimates of abundance are among the most important information needs of wildlife managers. Unfortunately, many carnivores are cryptic, secretive, and occur at low density. Accurate estimates of abundance are seldom obtainable for such species, so indices of relative abundance often substitute (see species accounts in Novak et al. [1]). Carnivore scent-station surveys are one such index.

We used standard methods to analyze scent-station data collected in Minnesota during 1986-93. Although our data set was among the largest in existence, we were frustrated by inadequate sample sizes. The most popular statistical model for scent-station data fit poorly. Anomalous data had undue influence on summary statistics and affected results of statistical comparisons. To overcome these problems, we devised improved methods for using scent-station surveys to monitor temporal and geographic trends in carnivore populations.

The difficulties we encountered can be traced to a few key features of survey designs and methods of analysis. These include the spatial distribution of scent stations, the experimental unit chosen for analyses, the statistic used to summarize results, the statistical model underlying analyses, and confounding of statistical comparisons. In this paper, we discuss these aspects of the design and analysis of carnivore scent-station surveys. Our presentation will demonstrate the use of field data to resolve issues raised in this paper.

Survey Methods

The carnivore survey conducted annually by the Minnesota Department of Natural Resources and the U.S. Fish and Wildlife Service was the source of field data for our presentation. Each scent station consisted of a 0.9-m diameter circle of smoothed earth with a scented lure placed at the center. Stations were grouped in lines to simplify data collection. Ten scent stations placed along an unpaved road at 480 m intervals comprised a line. Minimum spacing between lines was 5 km. Sampling was non-random, but 441 lines were distributed throughout the state. Each line was operated for one night each year between late August and mid-October, though not all lines were operated every year. Presence or absence of tracks was recorded, by species, at each station when it was checked the day after activation.

Choosing An Experimental Unit

Scent-station surveys vary in design. Sometimes stations are not grouped, as they were in Minnesota. The dispersion of stations should determine how stations are treated in analyses: in some cases, stations may reasonably be treated as independent samples; in others, they should be considered correlated samples or subsamples. Usually these issues are given inadequate consideration.

Closely spaced stations produce correlated data, but how far correlations extend is unknown. Stations placed too close to one another produce redundant data. Spacing stations more widely than necessary increases the cost of surveys and precludes intensive sampling of small areas. Subjective estimates of optimum spacing are inconsistent. Some investigators (e.g., Smith et al. [2]) have treated stations within 320 m of one another as independent samples. Others (e.g., Morrison et al. [3]) thought it necessary to separate stations by as much as 1.6 km. We have used variograms to show that correlations between stations often extend to 2000 m or more. Separating stations by this great a distance is seldom practical, so we have pursued the development of summary statistics and methods of analysis that are robust to correlations between stations.

Summary Statistics

Results of scent-station surveys are almost always summarized by visitation rates (ps= stations visited/stations operated). As a summary statistic, visitation rates have two serious deficiencies.

First, visitation rates are not directly related to abundance because each station has the capacity for only one detection. When visitation rates are high, many individual carnivores encounter stations that have already been visited. These additional visits have no effect on visitation rates. The result is a nonlinear relationship between visitation rate and abundance. The form of the curve is unknown, except for the y-intercept (0) and asymptote (y=1), so visitation rates can be used only to rank abundances.

Second, visitation rates are easily influenced by factors other than abundance, especially when sample sizes are small or visitation rates are low. These may include weather, season, human activity, or other factors that influence animal behavior. An ideal summary statistic would be robust to such effects. We will use examples to demonstrate the poor performance of visitation rates and present two alternative summary statistics: the proportion of lines that are visited (p1) and the negative natural logarithm of the proportion of lines that are not visited (-ln[1-p1]).

Statistical Models

For analytical convenience, some investigators treat stations as independent Bernoulli trials: a visit by one or more individuals of a species is a "success." This model leads naturally to convenient methods of analyzing binomial data, including logistic regression and log-linear models. The benefit of this approach is the ability to investigate variables that affect visitation probabilities of individual stations (e.g., habitat characteristics). Aggregating stations into groups--lines, in our example--and treating each group as an experimental unit is a more conservative approach. Group visitation rates are treated as independent random samples from unknown distributions. This approach has been used by investigators (e.g., Roughton and Sweeny [4]) who were unwilling to assume stations were independent.

To our knowledge, the fit of the binomial model has never been tested. We devised a goodness-of-fit test and found the binomial distribution to be a poor model for visitation rates, but an adequate one for the proportion of lines with one or more stations visited.

Statistical Comparisons

With few exceptions, statistical analyses of scent-station data have been limited to pairwise comparisons (e.g., of years, seasons, or geographic locations). Significant differences faithfully reflect changes in abundance only if other factors that affect visitation are relatively constant over time and through space. Some investigators are unaware of the possible confounding effect of other factors (e.g. weather). Most often, however, only two or three years of data are available and are inadequate for testing the significance of long-term trends. Long-term data sets and careful analysis are required for separating changes in abundance from changes in confounding factors. We advocate testing for trends by simple linear regression of rank-transformed data: the method is easy to apply and interpret and is robust to confounding.

Summary

Scent-station surveys are widely viewed as an accurate and inexpensive means of simultaneously gaining reliable information about the distribution and relative abundance of several species of carnivores (Johnson and Pelton [5]). Whether a particular scent-station survey will meet these high expectations depends largely on how the following issues are resolved:
  1. Sampling: How should stations be spatially distributed?
  2. Response variables: Is ps a suitable summary statistic?
  3. Statistical models: The binomial distribution has convenient properties, but does it adequately describe field data?
  4. Statistical comparisons: Are comparisons confounded by unidentified factors?

Acknowledgments

We thank the Minnesota Department of Natural Resources, especially W. E. Berg, and the U.S. Fish and Wildlife Service for generously providing survey data. Funding for manuscript preparation was provided by the Northern Prairie Science Center and the Wisconsin Cooperative Wildlife Research Unit of the Biological Resources Division, U.S. Geological Survey, and by the Graduate School, Department of Wildlife Ecology, and College of Agriculture and Life Sciences at the University of Wisconsin-Madison.

Literature Cited

  1. Novak, M., Baker, J.A., Obbard, M.E. and Malloch, B., eds. (1987) Wild furbearer management and conservation in North America. Ontario Trapper's Association, North Bay, 1150 pp.
  2. Smith, W.P., Borden, D.L. and Endres, K.M. (1994) Scent-station visits as an index to abundance of raccoons: an experimental manipulation. J Mammal 75, 637-647.
  3. Morrison, D.W., Edmunds, R.M., Linscombe, G. and Goertz, J.W. (1981) Evaluation of specific scent station variables in northcentral Louisiana. Proc Annu Conf of Southeast Assoc Fish and Wildl Agencies 35, 281-291.
  4. Roughton, R.D., and Sweeny. M.D. (1982) Refinements in scent-station methodology for assessing trends in carnivore populations. J Wildl Manage 46, 217-229.
  5. Johnson, K.G., and Pelton, M.R. (1981) A survey of procedures to determine relative abundance of furbearers in the southeastern United States. Proc Annu Conf Southeast Assoc Fish Wildl Agencies 35, 261-272.

Downloading Instructions -- Instructions on downloading and extracting files from this site.
(Download) carnivor.zip ( 12K) -- Carnivore Scent-station Surveys: Statistical Considerations
Installation: Extract all files and open index.htm in a web browser.

Accessibility FOIA Privacy Policies and Notices

Take Pride in America logo USA.gov logo U.S. Department of the Interior | U.S. Geological Survey
URL: http://www.npwrc.usgs.gov/resource/mammals/carnivor/index.htm
Page Contact Information: Webmaster
Page Last Modified: Saturday, 02-Feb-2013 05:45:59 EST
Reston, VA [vaww54]