OUP user menu

Accuracy of Scat Sampling for Carnivore Diet Analysis: Wolves in the Alps as a Case Study

Francesca Marucco, Daniel H. Pletscher, Luigi Boitani
DOI: http://dx.doi.org/10.1644/07-MAMM-A-005R3.1 665-673 First published online: 5 June 2008

Abstract

We assessed the accuracy of scat-sampling methods in relation to sources of bias (statistical independence of the data and definition of the sampling unit) and precision (sample size). We developed a method to quantify diets of predators accurately in a study of diet selection by wolves (Canis lupus) during 3 winter seasons (1999–2002) in the Western Alps. The best sampling design to avoid pseudoreplication was the “additive method,” where the presence of a carcass, estimated by either a collection of scats or a carcass itself along the travel route of a wolf, was considered 1 sampling unit. Although roe deer (Capreolus capreolus) were the primary prey used by wolves in the area, red deer (Cervus elaphus), recently reintroduced prey present at low density, were selected in winter 2001. We evaluated the optimal sample size for a given question using Monte Carlo simulations. At small sample sizes, slight increases in sample sizes caused large reductions in the standard error, greatly improving the precision of the estimates of percentage of items in the diet. Estimating the number of rare prey species used by wolves, such as red deer in our case study (<2% of the diet estimates), was possible if the minimum sample size was greater than 10–40% of the population of carcasses. We emphasized the importance of the additive method to improve the accuracy of estimates of diet selection by carnivores.

Key words
  • accuracy
  • Canis lupus
  • diet selection
  • Monte Carlo method
  • scat sampling
  • Western Alps
  • wolf

Fecal analysis to assess carnivore diets is widely used because the approach is inexpensive, relatively quick to apply, and large samples can be collected (Litvaitis 2000). The method is noninvasive and thus compatible with the endangered status of many carnivores in different countries (Mills 1996). Researchers have examined the diet of predators in Africa (Bailey 1993; Mills 1992) and of wolves (Canis lupus) in Europe (Capitani et al. 2004; Gade-Jorgensen and Stage-gaard 2000; Jȩdrzejewski et al. 2000) and North America (Kohira and Rexstad 1997; Marquard-Petersen 1998) using scat analyses. Scat analysis is useful to construct a basic description of a carnivore's diet, particularly when other methods are difficult or when one needs background information for planning broad-based studies of behavioral ecology (Mills 1996). Scat analysis is also a particularly useful supplement to radiotracking, direct observation, and snow-tracking in studies of diet selection.

Scat analysis has 3 main parts: scat sampling (collection), laboratory analysis, and data analysis. Although biases and interpretational difficulties have been considered for laboratory and data analyses (Ciucci et al. 1996; Reynolds and Aebischer 1991), sampling error, biases, and interpretational difficulties due to scat sampling have rarely been discussed.

Random sampling of scats in the field is rarely feasible, although the use of scat-detection dogs may improve this in the near future for some species (Smith et al. 2005; Wasser et al. 2004), potential biases may occur if nonindependent groups of scats collected at den sites, at kill sites, or along a track cause pseudoreplication (Hurlbert 1984) and overrepresentation of a prey species in the diet (Carss and Parkinson 1996; Huggard 1993b). In particular, predators with large prey relative to their size (e.g., wolf and weasel [Mustela]), can produce highly correlated clusters of scats. Conversely, solitary carnivores that eat generally small prey (e.g., coyote [Canis latrans]) produce fewer scats per individual prey item, which likely would lead to fewer problems with independence. Biases also can occur if nontarget scats are included in the analysis.

A sample of scats that is small compared with the total number of scats produced may be unrepresentative and subject to sampling error (Reynolds and Aebischer 1991; Trites and Joy 2005), especially to detect the presence of small or rare prey species in the carnivore's diet. If the sample size is too small, tests of hypotheses lack power (Bros and Cowell 1987; Sheppard 1999), especially in tests to compare diets (Reynolds and Aebischer 1991). Conversely, when the sample size is very large, the power of a specific test may be adequate, but effort may have been wasted in collecting and processing so many scats.

We developed and tested the “additive method” for improving accuracy in diet selection studies by supplementing data on wolf kills with data on scats collected through snow-tracking in a study of the diet of wolves in the Western Alps. The combined data of the additive method are more biologically meaningful than the data on scats alone because the sampling unit is a prey carcass itself and not a scat. With this approach, we could address the issue of independence related to groups of scats and avoid pseudoreplication.

Data on food habits and prey selection of wolves in the recently recolonized area of the Western Alps (Fabbri et al. 2007) may dispel misperceptions, help direct management decisions, and lead to more refined research hypotheses. Therefore, we investigated the significance of ungulates in the diet of recently settled wolves. Red deer (Cervus elaphus) were of particular management interest, because they were recently reintroduced and at low density. We evaluated the sample size necessary to document the number of red deer in the diet. We evaluated the optimal sample size using Monte Carlo simulations for the given question and we evaluated possible sampling errors. We discussed the importance of the additive method to improve the accuracy of estimates of carnivore diet selection and the applicability of the method to other carnivores under other conditions and seasons.

Materials and Methods

Study area.—The study area was defined by the territory of the “Valle Pesio” wolf pack and was located in a mountainous region of the Western Alps of Italy and France (Fig. 1). The area was approximately 500 km2 and encompassed Alta Valle Pesio Natural Park (67.7 km2) and adjacent lands. The core area was characterized by long, narrow valley bottoms surrounded by rugged mountains reaching altitudes of 800–2,651 m above sea level. Annual precipitation averaged 1,285 mm and the snow season lasted from November to April. Density of humans was 138 people/km2. The wild ungulate species in the area were chamois (Rupicapra rupicapra), roe deer (Capreolus capreolus), wild boar (Sus scrofa), and red deer. Roe deer, chamois, and wild boar were abundant as a consequence of reintroductions by the Italian and French Park Systems beginning in the 1980s and because of natural range expansion by ungulates throughout Italy (Mustoni et al. 2002). Red deer were reintroduced into the park in 1996.

Fig. 1

Study area defined by tracks of wolves and results from genetic analysis conducted on scats of the Valle Pesio pack followed during 3 winter seasons (1999–2002) in the Western Alps, Italy and France.

Scat sampling and analysis.—We searched for wolf tracks on skis or snowshoes using systematic transects that covered the entire area. When we found wolf tracks, we followed them and considered each continuous track a single travel route. We conducted winter scat sampling along travel routes of wolves during 3 winters from October 1999 to May 2002 (i.e., scats were collected along routes used by wolves, hence both off and on trails made by humans, far from and close to prey carcasses, etc.). We collected each wolf scat that we encountered, and froze it (#x2212;30°C) before analysis. We followed the laboratory procedures of Reynolds and Aebischer (1991). We identified mammalian hairs in each scat by microscopic examination of the cuticular pattern, the medulla, and the cross section. Accuracy of observers in identifying mammal hairs, evaluated through a preanalysis blind test on a sample of 120 hairs from local mammals, was 99% and consistency between observers was 97% based on reanalysis of random subsamples (Marucco 2003).

We estimated the diet of wolves using frequency of occurrence of the scat contents (Reynolds and Aebischer 1991). We generated 95% nonsimultaneous bootstrap confidence limits for the percentage of each food item in the diet. These limits represent the effects of random sampling error (Manly 1998). We simulated 2,000 sets for each bootstrap simulation.

Evaluation of precision and possible biases.—We assessed the accuracy of the scat-sampling method in relation to precision (sample size) for a given question and different sources of bias (inclusion of scats made by species other than wolves in the analysis, independence of the data, and the definition of a sampling unit).

We estimated the population of scats of the Valle Pesio wolf pack for the 3 winter seasons (1999–2002) based on the number of wolves present in the pack, a daily defecation rate (from 2 to 4 scats wolf−1 day−1), and a winter season of 181 days (November–April). We estimated this range of defecation rates based on a defecation rate of 3 or 4 scats wolf−1 day−1 for captive wolves fed ad libitum for short periods (Floyd et al. 1978; Weaver 1993) and considering that wild wolves eat variably, likely reducing the mean number of scats produced per day. We estimated the minimum number of pack members per month through snow-tracking and genetic results on scat samples for each winter season (Marucco 2003). Using the results from genetic analysis conducted on scats, we excluded from the analysis scats belonging to individuals of other packs. We estimated the precision of dietary estimates using an analysis of the sampling variance, which is measured by the squared standard error. We determined the sample size necessary for a chosen level of precision using a form of Monte Carlo simulation (Bros and Cowell 1987; Trites and Joy 2005). The Monte Carlo procedure we used to generate the standard error function was a sampling procedure from a realistic, finite population, generated using resampling without replacement. For each sample size, we randomly drew 1,000 samples from the estimated population of scats and computed the standard error. This number of resamples was a realistic minimum for a test of significance at α = 0.05 (Manly 1998).

Genetic analyses were conducted on a subsample of 150 scats by the Instituto Nazionale della Fauna Selvatica genetic laboratories in Bologna (Italy) to discriminate between different canid species (Randi et al. 2000). We calculated the proportion of scats made by species other than wolves present in the sample collection.

Independence of the datathe additive method.—We pooled wolf scats into “collections” defined as the aggregate of wolf scats collected along a continuous travel route used by wolves, uninterrupted by an identifiable carcass used or preyed upon by wolves, or by lack of snow. Prey carcasses were located when we backtracked wolves during the 3 winters of 1999–2002 (Huggard 1993a; Kunkel et al. 1999). Carcasses of prey found represented a known minimum number of prey consumed for the wolf pack. We supplemented the list of prey carcasses found during snow-tracking sessions with the data from scat collections. We assumed a missed carcass when collection of scats from a certain day contained hair of a species different from the preceding known kill or >3 days from the preceding known kill of that species (Huggard 1993b; Jȩdrzejewski et al. 2002). In a controlled feeding experiment, Floyd et al. (1978) found that wolves defecated undigested prey remains 8–56 h (0.3–2.3 days) after consumption. We estimated the date of defecation for each scat based on snow-tracking data for wolves and freshness of scat following Jȩdrzejewski et al. (2002). In this way, we reduced serial correlation on the contents and made the data on scats and carcasses additive. Hence, a prey carcass used by wolves, estimated by either a collection of scats or a carcass itself along a wolf travel route, was considered 1 sampling unit for this analysis. The description of the diet using this method was defined as the “additive estimate.” We used the same Monte Carlo procedure described above to estimate precision relative to different sample sizes of this new approach. We estimated the “carcass population” used by wolves in the area of the pack for the 3 winter seasons (1999–2002) based on the number of wolves present in the pack, the estimated live weights of prey (Mustoni et al. 2002), a winter season of 181 days (November–April), and a daily food consumption rate from 2.6 to 5.6 kg/wolf. We calculated the minimum food consumption rate based on the field metabolic rate (FMR) derived from Nagy's formula (1987): FMR (kJ/day) = 2.58W0 862, where W is body weight in grams (Glowacinski and Profus 1997). An adult wolf in Italy has an average body weight of 32 kg (Gazzola 2005). Energy requirements based on the field metabolic rate amounted to 19,727 kJ/day, which corresponds to 2.6 kg of meat per day for an adult wolf (Gazzola 2005). Therefore, we estimated a minimum food requirement per adult wolf of 2.6 kg and a maximum food requirement per adult wolf of 5.6 kg based on one of the highest estimates in Europe from Jȩdrzejewski et al. (2002).

We evaluated the sampling effort to look for wolf tracks and to follow tracks for collecting scats and kills for each winter season. We considered the effort as days of work for 1 person.

We tested the effectiveness of the additive method in removing autocorrelation in the data set using mixed-effects logistic regression models of clustered data, where scats were clustered within scat collections that were sampled along the same wolf track or at a kill site. We analyzed the presence or absence of each important prey species in each scat using year and abundance of prey as independent variables, and the code of the scat collection as the random factor. A comparison to an analysis using a fixed-effects logistic regression model, which ignored the clustering of the data, illustrated the importance of taking the clustering of the data into account. We used the MIXNO software for generalized linear mixed model analyses (Hedeker 1999).

We compared diet estimates evaluated using only data on scats, only data on kills, and the additive data set, using chi-square tests of independence.

Diet selection by wolves.—We compared use of prey evaluated by the additive method to availability of prey. Prey selectivity, the nonrandom representation of the available food in the observed diet (Chesson 1978), can be generated at several levels (Huggard 1993b). Here, we evaluated availability at the encounter level, where wolves already selected where to move because of spatial and environmental complexities. Prey availability was evaluated along systematic transects to estimate relative proportions of prey along travel routes used by wolves (Kunkel et al. 1999). We skied two 100-m transects in opposite directions and perpendicular to the travel route used by wolves at 1-km intervals. The location of the 1st transect was chosen randomly. The distance to the 1st tracks of roe deer, chamois, red deer, or wild boar on each transect was recorded; the distance recorded was 100 m if no track was encountered. The number of prey tracks located on both transects (0, 1, or 2; only the 1st track on each transect was recorded) was divided by the distance to that track (e.g., 1/190 if 1 roe deer track was found at 90 m in 1 direction and if no tracks were found in the opposite direction) to obtain the number of prey tracks per meter. This value was divided by the number of days since the most recent snowfall > 5 cm to adjust for snowfall effects (Kunkel et al. 1999). We did not ski transects after more than 7 days had elapsed since a snowfall because track deposition plateaued and tracks deteriorated after this time (Kunkel et al. 1999).

We calculated Manly's (1974) index α for each prey species (Krebs 1998) to estimate the dietary preference of the Valle Pesio wolf pack: Embedded Image

where ri is the proportion of prey species i in the diet (i = 1, 2, …, m), ni is the proportion of prey species i in the environment (i = 1, 2, …, m), and m is the number of prey species possible.

We normalized α values so that their sum equaled 1.0. Thus, if predation was not selective, α = 1/m; if a prey item was preferred, α > 1/m (Manly 1974). The additive method allowed us to estimate the standard errors of the α values following Manly (1974).

Results

Prey use from scat analysis.—We collected 435 scats along 694.1 km of wolf tracks (112 scats during winter 1999–2000 [winter 00], 179 during winter 2000–2001 [winter 01], and 144 during winter 2001–2002 [winter 02]). Using scat analysis alone, the winter diet of the Valle Pesio pack differed significantly among years (χ2 = 74.01, d.f. = 8, P < 0.001; Fig. 2). Wild ungulates (especially roe deer) comprised the majority of the diet during each winter (Fig. 2). Within species, red deer were more important in winter 01, whereas wild boar were more important during winter 02.

Fig. 2

Frequency of occurrence of food items in the diet estimated with data from only scats of the Valle Pesio wolf pack during 3 winter seasons (1999–2002), in the Western Alps of Italy and France, with 95% bootstrap nonsimultaneous confidence intervals for the proportion of each food item in the scats (n = 435). D.U. = domestic ungulates; S.M. = small mammals.

Evaluation of precision and possible biases.—The minimum number of wolves per month ranged from 2 to 8, with a mean (±SD) of 4.8 ±1.6 individuals per month (pack size was 5 in winter 00, 7 in winter 01, and 5 in winter 02). We estimated that the total population of scats ranged from 1,510 to 3,020 for winter 00, from 1,790 to 3,580 for winter 01, and from 1,186 to 2,372 scats for winter 02. The sample of 112 scats in winter 00, 179 scats in winter 01, and 144 scats in winter 02 represented 3.7–7.4%, 5.0–10.0%, and 6.1–12.1% of the respectively estimated populations of scats.

A larger sample size of scats reduced the standard error and increased the precision of the diet estimate based on data from scats alone for each species of wild ungulate prey (chamois, roe deer, red deer, and wild boar) during each winter (e.g., Fig. 3). If we considered sample sizes of scats < 200, the 95% confidence interval for red deer included 0 in winters 00 and 02 (Fig. 3). The standard error for each prey species declined rapidly until a sample size of approximately 200 scats was reached (Fig. 4). Therefore, a sample size > 6–16% of the population of scats should be collected to detect prey species that have a low percent of occurrence (<2%) in the diet estimate based on data from scats alone, as did red deer in winters 00 and 02. In our study we needed at least an effort of 200–250 days of work to accomplish this. In this way, the sample size is beyond the region of greatest change in slope of the standard error function (Figs. 3 and 4).

Fig. 3

Effect of sample size on 95% Monte Carlo confidence intervals (CIs) of the frequency of occurrence of red deer during winter 2000 in wolf scats of the Valle Pesio wolf pack in the Western Alps; the vertical axis is the proportion of wolf scats containing hairs of the prey species, for samples of increasing size (horizontal axis).

Fig. 4

Standard error (SE) of the occurrence of each prey species in the diet of wolves as a function of sample size of scats for A) winter 2000, B) winter 2001, and C) winter 2002, and as a function of sample size of prey carcasses used by wolves estimated by the additive method for D) winter 2000, E) winter 2001, and F) winter 2002, in the Western Alps.

Only 1 of 150 scats genetically tested did not belong to a wolf (it belonged to a fox). Inclusion of scats from species other than wolves therefore occurred too infrequently to bias the diet analysis.

Independence of the datathe additive method.—Through genetic analysis, we found that scats along a single travel route could belong to the same individual; for example, during winter 02, 4 consecutive scats along the same travel route were from wolf F3 (all scats contained chamois hairs), and 3 consecutive scats were from wolf M25 (all scats contained roe deer hairs). These scats were not independent samples for a study of diet selection by wolves.

We found 51 wolf kills and 4 scavenging events along 694.1 km of wolf tracks. Roe deer comprised the largest proportion of prey carcasses during every winter (winter 00: 0.91; winter 01: 0.69; and winter 02: 0.83; Table 1).

View this table:
Table 1

Proportion of the different ungulate species used by wolves (estimated by the additive method, using data on only kills, and data on only scats) with 95% bootstrap nonsimultaneous SE, in the Western Alps, Italy and France.

Mean number of scats per collection (± 1 SD) was 3.5 ± 2.5 (n = 46, range 1–10; winter 00: 2.6 ± 1.3, n = 16, range = 1–6; winter 01: 4.3 ± 2.9, n = 13, range = 1–10; winter 02: 3.8 ± 2.7, n = 17, range = 1–9). The additive method yielded a total of 27 prey carcasses in winter 00, 38 in winter 01, and 34 in winter 02 (Table 1).

We estimated a range of the total population of carcasses in the area used by wolves from 79 to 171 for winter 00, from 92 to 198 for winter 01, and from 46 to 98 for winter 02. The sample of 27 carcasses in winter 00, 38 in winter 01, and 34 in winter 02 represented 15.8–34.2%, 19.2–41.3%, and 34.7–73.9% of the respectively estimated population of carcasses.

A larger sample size of carcasses reduced the standard error and increased the precision of estimates of prey use for each species of wild ungulate prey (Fig. 4). The standard error for each prey species declined rapidly until a sample size of approximately 20 carcasses was reached (Fig. 4). Therefore, we should collect a sample size greater than 10–40% of the population of carcasses to detect red deer, which would require an effort of at least 200–250 days of work.

In the mixed regression models for clustered data (Table 2), the random term is included to account for the nonindependence of scats within collections. The random term, which describes the way in which scats from the same collection are similar and not independent relative to the sample as a whole, and how much covariance corresponds to this source of pseudoreplication, is highly significant for every species, except chamois (Table 2). Covariates, such as year and availability of prey, are of significant importance depending on the species of prey considered (Table 2). For each parameter of the model, maximum marginal likelihood estimates, standard errors, and P-values are provided. The value of –21og-likelihood (i.e., the deviance) at convergence is given; only the best model is reported and the relative fixed-effects model with no random effect (Table 2).

View this table:
Table 2

Results of the mixed-effect logistic regression models and logistic regression models with no random effects using the additive dataset separately for roe deer, red deer, chamois, and wild boar, collected during 3 winter seasons (1999–2002), in the Western Alps. For each parameter of the model, maximum marginal likelihood estimates, standard errors, and P-values are provided. The value of –21og-likelihood (i.e., the deviance) at convergence is given. These P-values are 2-tailed, except for the random effect variance where 1-tailed P-values are given.

We compared diet estimates evaluated with the 3 different datasets (only scats, only kills, and the additive data set; Table 1). Diet estimates analyzed using the additive method and using the data only from kills differed significantly (P < 0.05).

Prey selection.—Because 4 primary prey species occurred in our study area, α values of 0.25 would indicate that use reflected availability. Using the additive method, wolves of the Valle Pesio pack selected red deer (Manly's α = 0.44, SE = 0.12) over chamois and wild boar in winter 01, whereas roe deer were eaten as available (Fig. 5). In winter 02, wolves selected wild boar (Manly's α = 0.51, SE = 0.11) over chamois and red deer, whereas roe deer were eaten as available (Fig. 5).

Fig. 5

Prey selection by wolves of the Valle Pesio pack using Manly's α values (error bars are 1 SE), in the Western Alps, Italy and France. Prey use is evaluated by the additive method and availability of prey (relative proportions) by counting tracks of prey species along travel routes used by wolves.

Discussion

An estimate of the diet is often the 1st step to providing the necessary background for more-detailed, broad-based studies of predator–prey ecology (Mills 1996), but the limitations of the fecal method must be borne in mind. The procedure we adopted provides a tool for improving and evaluating the accuracy of the sampling design for collecting scats for specific questions in studies of food habits of wolves and other carnivores. Strong inference can be achieved with an accurate sampling design, reduction of bias, and increasing precision (Manly 1996; Sheppard 1999). Combining data sets from scat collections with data collected on kills can improve the accuracy of estimates of prey selection; however, this approach may lead to erroneous results if the 2 sets of data are not compatible and not accurate (Mills 1996).

Bias.—Bias and misleading conclusions associated with scat collection can result from the inclusion of scats from species other than wolves in the analysis, nonindependence of data, and an incorrect definition of the sampling unit. A conservative, multicriteria approach to differentiate wolf scats from those of other canids is usually used (Ciucci et al. 1996), but bias may still occur due to including scats from other species or due to discarding some wolf scats. Using genetic analysis, we detected only 1 scat from a fox out of 150 scat samples. In areas where wolf and dog scats occur, the issue of incorrectly including scats from other species should be regularly evaluated by appropriate genetic sampling.

Potential biases may occur if sampling scats at rendezvous sites, kill sites, or along a track or a trail causes overrepre-sentation of a prey species in the diet (Marquard-Petersen 1998; Scott and Shackleton 1980) because of nonindependence of scat samples. Theberge et al. (1978) found higher proportions of beavers (Castor canadensis) in scats from rendezvous sites compared with collections from other areas used by wolves in Algonquin Park, Ontario, Canada. Clusters of scat collected at a kill site also can overestimate that prey species in the diet. Mattson et al. (1991) subsampled scats where more than 5 were found at 1 kill site. In our study, we found that clusters of scats collected on the same day, along the same wolf track, or at a kill site, that contained hair from the same kill were not independent; therefore, we considered a collection of scats and not a single scat as a sampling unit to avoid pseudoreplication (Hurlbert 1984).

The additive method.—The use of scat analysis may be important to document diet selection or estimate consumption or kill rates of carnivores, if the data set is combined with data from kills. This final data set was more biologically meaningful than the data from scats alone because the sampling unit was the prey carcass itself and not a scat. The additive method is optimal in studies of diet selection because it deals with the issue of independence, which is fundamental for studies of selection (Thomas and Taylor 1990). The mixed-effect analysis indicated the importance of considering collections and not single scats as sampling units to minimize the problem of nonindependence of scats. The additive method also allows researchers to estimate the standard error of Manly's index of selection (1974). The same approach can be adopted in the summer, and in areas where snow is absent, if other techniques, such as direct observations or radiotracking, are conducted to collect data on kills.

In our study, the additive data of carcasses and scat collections was more representative than the data based only on wolf kills. It represented a higher percentage of the total population of carcass samples present in that area at that period of time. In fact, chamois was underrepresented in analyses only of wolf kills; this apparent avoidance of chamois was likely an artifact of the difficulty in documenting chamois carcasses in rugged terrain. This problem was taken into account by adding results from scat collections in the final additive data set. Jȩdrzejewski et al. (2002) recommended the supplementation of analyses of scat contents with data on kills in studies of wolf predation in dense woodlands and in regions where wolves consume small and medium-sized ungulates (roe deer and piglets of wild boar). The additive method may still underestimate small prey species if ≥2 individuals of the same species of prey, killed and consumed by wolves in a short time, are counted as 1 prey item if only recovered from scat collections. Moreover, it is not possible to differentiate between scavenged and killed prey with scat analysis; therefore, prey consumption and not necessarily predation is determined.

Precision.—Roe deer comprised the majority of the estimated diet of wolves during each winter. Red deer and wild boar were less abundant in the additive estimate but were selected respectively in winters 01 and 02, suggesting prey switching. Domestic ungulates, mainly goats and sheep, appeared in the estimated diet of wolves in winter when data only from scats were used, as a consequence of livestock depredations during the summer. The presence of livestock in the diet in winter was the result of “food caching” (Mech 1970) or scavenging because no livestock were available for the wolves during the winter season. For this reason we did not consider domestic ungulates in the diet selection analysis.

We compared the power of various sample sizes to detect each food item in the diet estimates, and the relationship was a decreasing asymptotic function approaching 0. The Monte Carlo procedure we used allows estimating the standard error of a statistic, using repeated samples from the original data set (Manly 1998). To deal with a finite population, such as the population of scats from a wolf pack, we obtained good estimates of precision by estimating the standard error using repeated samples from the original data set without replacement, using this form of Monte Carlo simulation (Manly 1996). With samples from the highly nonnormal distribution that are often encountered in biological studies, the method has the potential to be the most useful available approach for determining sample sizes (Manly 1998).

At small sample sizes of scats and carcasses, slight increases in sample sizes caused large reductions in the standard error, whereas at large sample sizes, further increases in sample size did not greatly affect precision. In particular, to precisely document common prey species in the diet (such as roe deer in our case study), a large sample is not needed (<5% of the population of scats and <10% of the population of carcasses). Red deer was the species selected in winter 2001. Red deer represented a small proportion of the diet of the Valle Pesio wolf pack in winter, but were less abundant and encountered less frequently in the study area than other ungulate species. Despite their low availability, red deer were selected during the severe winter of 2001 when the larger wolf pack was present (6–8 individuals). Okarma et al. (1995) documented a significant positive correlation between pack size and the number of red deer killed in Poland. To detect red deer using the additive estimate, we needed a minimum sample size greater than 10–40% of the population of carcasses. Using only data from scats, we needed a sample greater than 6–16% of the population of scats. The collection of the additive data set did not require additional effort, and for both data sets we needed at least 200–250 days of work in a winter for red deer. Predation on species that occur at a low percentage in the diet can be important to estimate if the prey species are selected by predators but have low availability, because predation could have a high impact on such species; and if the uncommon prey species is of high management interest. In our case study, we determined that this analysis for red deer is not worth the effort in the future. Instead, an in-depth study focused on population dynamics and behavior of red deer may be warranted to evaluate the impact of wolf predation on this species.

To detect differences in the diet between packs (or years), or to answer other questions, a researcher should evaluate the sample size of scats or carcasses necessary for a specified level of reliability, taking into account the size of the population of scats or carcasses (related to the number of wolves present in a pack, to the number of packs monitored, to the defecation rate or kill rate of wolves, and to the time period considered). We suggest the use of this Monte Carlo approach. This approach is fundamental for comparing and interpreting results among studies of diets.

Applicability.—Simultaneous examination of sampling design, power, and sampling effort allowed evaluating the costs and benefits of using the additive approach in a study of diet selection by wolves. By using this procedure in other studies of diets of carnivores, it is possible to evaluate the trade-offs between increasing sample size and increasing effort (depending on a specific research question), and at the same time avoid pseudoreplication. The additive method will be useful for application to most large carnivores with large prey relative to their size (e.g., lion [Panthera leo] and wild dog [Lycaon pictus]), and also to solitary carnivores that spend considerable amounts of time at kill sites (e.g., European lynx [Lynx lynx]— Molinari et al. 2000), producing highly correlated clusters of scats. Carnivores that generally consume prey with smaller body sizes likely produce fewer scats per individual prey item, which likely will not result in problems of independence. On the other hand, it will be difficult to document these kills, if the prey is so small that it will be consumed entirely in a short time. In these cases, combining analyses of scat contents with data on kills will allow a better representation of the missed kills not documented by other techniques. Therefore, the additive method can be widely applied and will be useful in a variety of situations.

Acknowledgments

Funding for this project was provided by the European Community and the Piemonte Region. We thank Alpi Marittime Natural Park and Alta Valle Pesio Natural Park, Italy, for logistical support and fieldwork. Special thanks are extended to the staff of the Progetto Lupo Piemonte for fieldwork and laboratory analysis. Genetic analyses were conducted by the Istituto Nazionale della Fauna Selvática, in Bologna, Italy. We are very grateful to R. A. Powell, E. Revilla, and an anonymous reviewer for their valuable comments and critical reading of the manuscript.

Footnotes

  • Associate Editor was Roger A. Powell.

Literature Cited

View Abstract