Northern Prairie Wildlife Research Center
Effective tests for spatial dependence require large datasets, especially when visitation rates are low. We have not encountered such tests in the scent-station literature. Our analysis of data collected in Minnesota revealed spatial correlations between stations that extended to approximately 2,000 m. Effects of station spacing were especially strong for mobile habitat generalists that defend territories to varying degrees (wolves, coyotes, red foxes, bobcats), and were much weaker for less-mobile species that display stronger habitat preferences (skunks, raccoons). Thus, multiple visits by individuals are the probable cause of spatial correlations.
|Fig. 4. Distribution of Spearman's rank correlations between population size and visitation rate for 10,000 sets of simulated binomial data directly relating animal abundance to visitation rate. Data were modeled after scent-station surveys of raccoons conducted by Smith et al. (1994). Correlations within the shaded region are significant at P ≤ 0.05 (1-tailed test).|
In Minnesota, spatial correlations had 2 practical consequences. First, statistical tests detected differences more often than they should have when we treated stations as independent experimental units. Second, a few lines with many visits had disproportionate influence on conclusions for some species, years, and sections. We countered the disproportionate influence of these lines by dichotomizing the result for each line (no visit, ≥1 visit). The resulting line index was robust to effects of nonrandom line placement, behavioral differences among carnivores, and observer error. Whereas the number of stations visited had an intractable statistical distribution, the number of lines visited was a binomial variable. These benefits outweighed the only disadvantage: a possible reduction of statistical power. Other means of reducing the weight afforded to multiple visits (e.g., the square-root transformation) produced intractable data-expressed in inconvenient units of measure.
Instead of comparing time periods, we used regression of rank-transformed data to test for trends. The method provided an approximate test for homogeneity of trends among sections and a general test for sustained trend of any form (e.g., linear, quadratic, or logarithmic). When the null hypothesis was rejected for a section, the observed ordering of survey results was unlikely to have occurred in the absence of an increasing or decreasing trend in visitation. Whereas differences in visitation between time periods can result from changes in factors other than abundance, such factors fluctuate irregularly from year to year and are comparatively unlikely to cause significant trends in visitation. To be detectable, trends must be large in comparison with annual fluctuations, and hence of significance to wildlife managers. In short, regression of rank-transformed data is simple to apply, robust to spurious conclusions, easy to interpret, and readily detected trends in our data. The principal challenge was choosing a time span for testing. Monotonic trends across years do not persist indefinitely, but analyses of very short spans are susceptible to confounding and have low statistical power.
Linhart and Knowlton (1975) and Roughton and Sweeny (1982) cautioned against using scent stations to compare carnivore abundance among areas of dissimilar habitat because differences in sampling biases and differences in carnivore abundance may both affect survey results. Unlike temporal biases, confounding effects of habitat differences persist through time. Unless they can be accounted for, spatial comparisons can easily lead to spurious conclusions. Thus, we did not attempt spatial comparisons of visitation rates. However, we did test the consistency of trends across sections and detected differences that document problems with pooling data from different habitats or geographic regions. In Minnesota, strong trends in a few areas contributed disproportionately to statewide trends in line indices for skunks and raccoons. Important local trends may, in some cases, cancel one another. Analyses of statewide trends for all species would have sacrificed resolution that is useful to managers.
|Fig. 5. Distribution of P-values obtained for 10,000 logistic analyses of simulated binomial data directly relating abundance to visitation rate. Data were modeled after scent-station surveys of raccoons conducted by Smith et al. (1994). P-values <0.05 are shaded.|
Our reevaluation of Smith et al. (1994) emphasized the role sample size plays in the outcome of validation experiments. The stated purpose of that study was to test the null hypothesis that visitation rates of raccoons were not independent of population density. The null hypothesis actually tested, however, was the opposite. After failing to detect a relation between visitation and abundance, the authors concluded that changes in density may precipitate changes in behavior that preclude the use of scent stations to index raccoon abundance. Our simulations implicate low statistical power as a more parsimonious and equally convincing explanation for results obtained by Smith et al. (1994), even though we probably overestimated the power of their tests (simulated data were not affected by errors in estimates of population size, seasonal variation in visitation rates, or observer error). Thus, it is appropriate to conclude the survey was not a useful index to raccoon abundance on Davies Island because it was deficient in statistical power. A more complicated biological explanation is unnecessary, and results of Smith et al. (1994) should not be generalized to other situations.
Our reevaluation of Diefenbach et al. (1994) highlights the importance of including experimental controls when designing validation experiments. Factors that influence carnivore movements affect the rate at which carnivores encounter stations. The motivation for carnivores to investigate attractants may include such factors as curiosity, hunger, or sexual interest, and these factors must compete with wariness toward attractants. Thus, weather (Leberg et al. 1983, Nottingham et al. 1989), season (Griffith et al. 1981, Smith et al. 1994), habitat characteristics (Linhart and Knowlton 1975, LeCount 1982, Nottingham et al. 1989), and human activity (Griffith et al. 1981, Andelt et al. 1985) are thought to affect visitation rates. The experimental design of Diefenbach et al. (1994) did not permit effects of likely confounders to be distinguished from effects of abundance. Effects of confounders were unambiguously suggested by our results, but effects of abundance were not.
The general validity of scent-station indices has been neither proven nor called into serious question by objective validation experiments. Moreover, we believe logistical constraints will preclude conclusive, experimental validation in many settings and for many species. Powerful experiments require estimation and manipulation of carnivore populations over a larger area than was feasible for Smith et al. (1994; W. P. Smith, U.S. Forest Service, personal communication). Similar practical considerations precluded experimental control at Cumberland Island (D. R. Diefenbach, Pennsylvania Game Commission, personal communication). Yet, we chose these 2 examples for review because they are among the largest, most sophisticated validation studies that have been conducted.
|Fig. 6. Visitation rates for bobcats to scent stations on Cumberland Island, Georgia, after (A) 11 September 1989, and before (B) 28 February 1989, and (Diefenbach et al. 1994).|
However, available evidence supports use of scent stations only for monitoring broad temporal trends in relative abundance at an intermediate scale of spatial resolution (e.g., by section). The method is ill-suited for monitoring species that are rarely detected, and for localized monitoring of wide-ranging carnivores. Many studies have used only a few dozen stations per survey, but reliable results may require hundreds of lines of many stations (Zielinski and Stauffer 1996), especially for species detected infrequently. Hence, the perception that scent-station surveys are cost-effective may not reflect the cost of obtaining a useful number of samples.
To obtain reliable results, investigators must use analyses that accommodate statistical properties of scent-station data. Individual carnivores respond differently to stations and are not equally detectable. Repeated sampling of the same individual is pseudoreplication and affords undue influence to individuals that visit many stations in succession. Lines are an appropriate experimental unit. Dichotomizing results for lines produces data with a tractable statistical distribution and affords equitable influence to individual carnivores.
Finally, scent-station indices cannot yet be converted to estimates of abundance. Thus, it is difficult to determine from survey results when management responses are warranted. Scent stations can be used to help identify trends, but should supplement, rather than replace, information from other sources. In this supplementary capacity, scent-station surveys are likely to remain a useful tool for carnivore research and management.