We studied size-selective predation by Common Eiders (Somateria mollissima) feeding on blue mussels (Mytilus edulis). Selection varied with location, season, and prey availability, and for the most part ducks preferred smaller mussels than those that would have been the most energetically profitable on a per prey basis. We tested predictions from two related hypotheses concerning optimal prey selection: (1) the shell-mass minimization hypothesis, which states that predators select food that minimizes shell ingestion, as opposed to maximizing energy intake per prey item; and (2) the risk-averse foraging hypothesis, which, assuming large prey are not preferred (because of high shell content, low energy value, or some other reason), states that predators feed on smaller prey when the risk of mistakenly taking large prey increases. We found support for both hypotheses, although the risk-averse foraging hypothesis should be tested further due to conflicting results and small sample sizes. During most of the year, Common Eiders selected relatively small mussels that minimized shell ingestion, even though larger available prey would have provided higher net energy gain per prey item. In winter, differences among length classes in shell ingestion became small, and birds switched to feeding on larger prey that provided more energy per unit work and probably were more profitable. During times when small and mid-sized prey were preferred, ducks foraging where large mussels were abundant usually selected smaller length classes of prey than did those feeding where large mussels were less common. Birds avoided the risk of inadvertently ingesting large prey by selecting smaller mussels. Our results provide insight into the mechanisms of prey selection by Common Eiders and may also help explain some of the discrepancies reported in previous studies of prey-size selection in this species. Received 31 October 1997, accepted 26 August 1998.

Department of Zoology, University of Guelph, Guelph, Ontario N1 G 2 W1, Canada MANY OPTIMAL DIET MODELS assume that an- imals sample a variety of prey types or sizes such that energy intake is maximized relative to costs associated with feeding (Stephens and Krebs 1986). However, the performance of such models under complex natural situations is questionable (Ball 1994), and it is seldom pos- sible to determine how animals perceive costs, benefits, and risks associated with foraging choices. The ability of individuals to assess their environment and distinguish among prey may influence how they perceive these factors, and in turn, may determine the degree to which they are selective and the prey that they choose (e.g. Elner and Hughes 1978, Hughes 1979). It might be inferred, from the observation Present address: Department of Biology, Univer- sity of New Brunswick, P.O. Bag Service 45111, Fred- ericton, New Brunswick E3B 6El, Canada. E-mail: dhamilt2@unbsj.ca 2 Address correspondence to this author. E-mail: tnudds@uoguelph.ca that prey are not always selected according to model predictions, that animals feed "subop- timally" (e.g. Draulans 1984). However, these animals may be feeding in a manner that max- imizes net reward conditional on constraints they face and their ability to distinguish prey. For example, Ball (1994) argued that animals foraging in highly variable environments where prey availability and profitability (net energy gain per search and handling time) change unpredictably, or where they face a wide array of choices, may be unable to distin- guish small differences in profitabilit3 and in- stead use "rules of thumb." Use of such rules may cause individuals to feed in a manner that appears to be suboptimal (i.e. not choosing the most profitable prey). On the other hand, if prof- itability changes predictably (e.g. seasonally), animals may be better able to assess the envi- ronment and feed more selectively. Considerable research has been conducted on prey selection by aquatic birds (e.g. Drau- lans 1982, 1984, 1987; Meire and Ervynck 1986; Bustnes and Erikstad 1990; Ward 1991; De- Leeuw and Van Eerden 1992; Ball 1994; Barras et al. 1996). Birds often eat prey that appear to be of low quality. Hypotheses to explain this include: (1) individuals minimize shell inges- tion (Bustnes and Erikstad 1990) or salt intake (Nystr6m et al. 1991) as opposed to maximiz- ing short-term energy intake; (2) large prey are more costly in terms of handling time and crushing resistance and therefore should be avoided, even if energetically rich (i.e. they are less energetically profitable; DeLeeuw and Van Eerden 1992); and (3) predators are unable to discriminate among prey of different sizes (Ward 1991). Hence, when attempting to deter- mine the optimal prey for foraging animals, re- searchers should consider more than just en- ergy return per prey item. Draulans (1982, 1984) suggested that ducks select mussels of smaller than optimal size be- cause this enables them to avoid the risk of in- gesting one that is too large to handle. Drau- lans' (1982, 1984) hypothesis assumes that ducks are imperfect at distinguishing prey of different sizes; the better their selection skills, the less they should have to compensate in higher-risk situations (Draulans 1987). The "risk" referred to by Draulans concerns the consequences of selecting a prey item that can- not be eaten. The bird has expended energy to find and handle the prey, but obtains no re- ward. In this situation, where predators cannot be certain of distinguishing these risky prey from the more beneficial ones (assume, for ex- ample, that the best prey are of intermediate size), the best strategy may be to select much smaller prey. These prey items may not provide as much benefit as preferred prey, but they pro- vide some benefit and are sufficiently different from large (i.e. risky) prey that foragers will not make a mistake. This idea may be extended beyond the case where large prey provide no benefit to situations where they are simply sub- stantially less beneficial (because of reduced energetic profitability, increased shell content, or some other factor). We examined predation on blue mussels (My- tilus edulis) by Common Eiders (Somateria mol- lissima) in two areas of Passamaquoddy Bay, New Brunswick, Canada. Common Eiders are present year-round in the area, and adults feed primarily on mussels (Hamilton 1997). Using a series of experiments, we attempted to deter- mine whether these ducks were size-selective predators, and if so, what mechanisms they used to choose their prey (i.e. what aspect of prey quality they selected for). We tested two hypotheses, each of which per- tains to a different aspect of optimal prey choice. The shell-mass minimization hypothe- sis (Bustnes and Erikstad 1990) posits that Common Eiders select prey items that mini- mize shell intake rather than maximize short- term energy gain. Under this hypothesis, the most beneficial prey are those with the smallest shell content but not necessarily the highest en- ergetic content. Relative tissue mass (tissue rel- ative to shell) is highest in small mussels (Bust- nes and Erikstad 1990), but differences among sizes may vary with season. At times when these differences are great, birds should choose relatively small prey, but when tissue content is more similar among mussel sizes, other selec- tion factors (i.e. energetic profitability) may come into play. Under the risk-averse foraging hypothesis (Draulans 1982, 1984), ducks select small prey that reduce the risk of taking prey too large to be handled (or that are unfavorable for some other reason). Although we are un- aware of any studies that examine the ability of Common Eiders to distinguish among length classes of prey, other species of diving ducks (Draulans 1982, 1984, 1987; Ball 1994), as well as other birds (Ward 1991), are known to be im- perfect at separating similar sizes. Assuming that Common Eiders are capable of some size discrimination, but not perfect at distinguish- ing prey of different sizes, in areas and seasons where large mussels are less beneficial (possi- bly because of high shell content) and more abundant, ducks should switch to feeding on smaller prey than they would under conditions where large prey are less common. We also used our experiments to attempt to explain var- iation in prey sizes selected by Common Eiders in several previous studies (Raffaelli et al. 1990, Nystr6m et al. 1991, Guillemette et al. 1996). METHODS Field experiments.--We performed prey-selection experiments at two locations (Barr Road and Indian Point) approximately 3 km apart near St. Andrews, New Brunswick, Canada (4504'N, 67 02'W) during spring (April and May), summer (June through Au- gust), and fall (September and October) 1995, and winter (January through March) and spring (April and May) 1996. The study area remained ice free throughout the experiment. Our study included for- aging by only adult and full-grown juvenile (in fall) ducks. Sex ratios were approximately equal except in summer, when many females were feeding else- where with ducklings. Duckling feeding grounds were excluded, because ducklings and associated fe- males have a substantially different diet than adults without ducklings (Cantin et al. 1974, Hamilton 1997). The population structure was similar in the two study locations. Analyses of ducks collected from the study areas suggest that blue mussels were the most commonly eaten prey for adult Common Ei- ders in all seasons (Hamilton 1997). Blue mussels were collected from these areas and placed in aquaria with a flow-through seawater sys- tem in a laboratory at Huntsman Marine Science Centre. Square ceramic floor tiles (900 cm 2) served as a substrate for mussel attachment in the lab. Mussels were divided into four length classes: (1) 10 to 19 ram, (2) 19 to 28 mm, (3) 28 to 37 mm, and (4) 37 to 50 mm. All classes were within the range ducks were physically capable of ingesting (Hamilton 1997). We divided tiles into four equal sections using corru- gated plastic dividers (height ca. 3 to 4 cm) attached with silicone. Two types of tiles wer. e set up (Fig. 1). "Regular" tiles were those in which each section contained a different length class of mussels (i.e. one class per section). The surface area covered by mus- sels in each section was held constant so that ducks did not perceive differential mussel cover (145 mus- sels in class one, 66 in class two, 40 in class three, and 21 in class four). "Manipulated" tiles had six large mussels added to each section containing one of the three smaller length classes. In order to maintain constant surface area both within manipulated tiles and between manipulated and regular tiles, we also reduced the number of mussels in the three smaller length classes on manipulated tiles (110 in class one, 52 in class two, 30 in class three, and 21 in class four). For both types of tiles, we randomized the location of sections containing different length classes on each tile. Use of two tile types allowed us to compare prey selection under conditions of equal cover of all length classes with choices made when large prey were more abundant. This provided a direct test of the risk-averse foraging hypothesis. We placed tiles in aquaria with aeration and a flow-through seawater system for three to seven days, until mussels attached by means of byssal threads. Aquaria were drained and left empty for ap- proximately 2 h daily to simulate a tidal cycle and to accelerate attachment (VanWinkle 1970). Mussel at- tachment was checked by turning tiles upside down. After attachment was complete, we recounted mus- sels in each tile section (sometimes a few mussels died or failed to attach). We placed tiles (in groups of three to six) in the intertidal zone in areas where ducks were known to Regular tile 30 cm 30 cm Manipulated tile 30 cm 30 cm FIG. 1. Diagram of typical regular and manipu- lated tiles with mussels of the four length classes. The number of mussels in each section does not re- flect the actual number used in experiments (see text). Positions of the four size classes were random- ized for each tile. feed. Distance between tiles within groups varied, but all were usually within 50 to 75 m of each other. Locations of groups of tiles set out at each study area depended on where ducks had last been observed feeding. At Indian Point, tiles were placed along a 2- km length of beach, whereas at Barr Road the study area was approximately 1 km long. All tiles were at about the same depth (5.5 to 7 m underwater at high tide, depending on the moon phase) to reduce po- tential effects of diving depth on prey selection (Draulans 1982, Beauchamp et al. 1992, DeLeeuw and Van Eerden 1992). Tiles were near the low-tide line and were exposed for a very short time each day. This minimized possible predation by gulls (which were common) and crows (which were much less common), because they did not feed underwater. However, we also placed several tiles higher in the intertidal zone and observed them using a spotting scope to determine whether gulls attempted to re- move mussels from these tiles. In all cases gulls avoided the tiles, so we concluded that their possible effects as predators were minimal. We placed a single tile from each group under a predator-exclusion cage designed to prevent ducks from feeding under it. These cages were 30 cm high, with a roof made of 3 x 3 cm plastic mesh, and had no sides. In most of the Indian Point study area, cag- es were permanently anchored structures of 1.5 x 1.5 m. These exclosures were used in conjunction with a predator-exclusion experiment (Hamilton 1997) and have been shown to effectively exclude ducks (Ham- ilton et al. 1994). At Barr Road and a small part of the Indian Point site (when tiles were positioned far from the permanent cages), a portable cage, approx- imately 75 x 75 cm, was used. Cages did not attract or exclude other predators and offered no protection from wave action (Hamilton 1997). Cages therefore acted as controls for mussels lost due to wave action or other predators (e.g. crabs and seastars) that also feed on mussels in the area. The different sizes of the two cage types almost certainly did not influence our results because observations of effects of duck exclu- sion on the mussel bed suggested that birds avoided cages completely, not even reaching under the edges of them in an attempt to retrieve mussels (D. Ham- ilton pers. obs.). Control tiles were located among ex- perimental tiles with the control associated with each group never more than about 40 m from any tile in the group. We used a spotting scope to check for ducks feed- ing in areas where tiles were positioned. Tiles were checked daily at low tide. When it became clear that mussels were missing from tiles in at least one length class (or within three days if no mussels were miss- ing), we recovered tiles and removed and counted the remaining mussels. The number of missing mus- sels that could be attributed to duck predation was determined by subtracting the number of mussels in each length class missing from the control (protect- ed) tiles from those missing from the experimental tiles that were exposed to predation during the same time period. We estimated the natural availability of mussels (hereafter "ambient") of different length classes by collecting all mussels from 20 100-cm 2 samples in each area during summer 1995. Mussels from each sample were counted and classified into length clas- ses. We obtained size-frequency distributions of mussels for Indian Point in the remainder of 1995 and 1996 using data collected from another experi- ment (Hamilton 1997). We also periodically (July, August, and December 1995; March and May 1996) collected sets of approximately 40 mussels (10 to 50 mm long) from the study area to assess relative tis- sue and shell mass at different times of the year. Mus- sels were opened and the tissue and shell dried sep- arately for 20 h at 90C, then weighed. We measured shell thickness (at the top of the valve near the at- tachment site of the posterior adductor muscle) and crushing resistance during one collection period. Force (N) required to crack mussels of different lengths was assessed using a Hounsfield tensometer. Mussels were placed width wise in the tensometer with one valve in contact with each of the crushing surfaces. Tension was increased slowly and the force at which mussels first cracked was recorded as crushing resistance. We assessed benefits of different prey-length clas- ses for ducks using several means. Energy (J) gained and shell mass consumed by eating mussels of each length were estimated for all seasons using estimates of energy content of mussel tissue (1 g dry tissue = 20.511 kJ; Bustnes and Erikstad 1990) and predicted average shell mass for each mussel length class. As estimates of costs associated with feeding on differ- ent sized mussels, we calculated the amount of work (in J; force x distance compressed) required to crack mussels of each length, and we determined force re- quired to pull mussels off tiles using a Pesola scale. From literature estimates of daily food requirements for Common Eiders (Bdard et al. 1980, Bustnes and Erikstad 1990, Egerrup and Laursen 1992, Hilgerloh 1997), we estimated an average dry-tissue biomass requirement of 130 g per day. We then calculated for each season the mass of shell that would be con- sumed per day from each length class if a Common Eider obtained the 130-g requirement completely from mussels of that length. The estimate of 130 g per day is likely to be an overestimate in summer and an underestimate in winter (Hilgerloh 1997), but this seasonal variation does not affect our interpretation because the relevant comparisons of shell ingestion are among length classes within seasons, not across seasons (see Results). We set the constant require- ment at 130 g per day simply to provide a uniform graphical presentation; any value would have given the same result. Statistical analyses.--We used a series of chi-square analyses to test for size selectivity (based on mussel length) at different times, locations, and among tile types (regular vs. manipulated). Each treatment combination was analyzed separately using a single calculated X 2 , although when insufficient replication was available, some seasons had to be pooled (e.g. spring and summer in a particular location). We used the following procedure to calculate X 2 values for each combination of time, location, and tile type. For each tile within a treatment combination, corrected (for losses other than to ducks) numbers of mussels eaten from each length class were taken as observed values, and expected numbers eaten were calculated based on a null hypothesis of random removal. We TABLE 1. Regression equations predicting dry-tissue mass from mussel length for the different months in which mussels were sampled. The relationship is log-log taking the form: tissue mass = 10 x length '. Month b a r 2 df F P July 1995 2.92 -5.36 0.97 1, 82 2,543.7 <0.0001 August 1995 2.53 4.80 0.96 1, 41 1,135.0 <0.0001 December 1995 2.67 -5.00 0.97 1, 39 1,172.9 <0.0001 March 1996 2.78 -5.23 0.98 1, 39 1,920.5 <0.0001 May 1996 2.45 -4.63 0.94 1, 39 660.3 <0.0001 summed observed and expected numbers across all tiles in an experimental group to generate single ob- served and expected values for each length class, and an overall X 2 value. We used this particular compu- tational approach because Common Eiders selected mussels at the level of the individual tile. However, because approximately the same number of mussels was available on each tile, we could equally have summed all observed values and calculated expec- tations based on those totals without changing the results appreciably. We then tested for differences in prey selection by Common Eiders among times, lo- cations, and tile types using a series of heterogeneity X2 analyses (Zar 1996). This procedure compared de- viation of observed from expected values (i.e. results of the simple X2 analysis) among different experi- mental manipulations (e.g. regular versus manipu- lated tiles in a given season and location) and was approximately analogous to testing for an interac- tion between main effects in a two-factor ANOVA (see Zar [1996] for computational details). All X2 val- ues reported in the text are from heterogeneity anal- yses. This nonparametric analysis was necessary be- cause when data were converted to a form that would accommodate ANOVA or other parametric techniques, assumptions of normality and homoge- neity of variances were severely violated. We pooled tiles within treatments because in many cases too few mussels were removed, resulting in violation of chi-square assumptions if each was considered sep- arately (Zar 1996). Our approach was conservative because it resulted in reduced degrees of freedom and potentially minimized observed effects by com- bining observations (different tiles) taken under slightly different conditions. It did, however, provide an estimate of overall prey selection by Common Ei- ders over the range of conditions they encountered. Our approach also had the advantage of giving greater weight to tiles on which more predation had occurred. All parametric analyses were performed using SAS version 6.11. Data were examined for confor- mity to assumptions and, when necessary, were transformed. We determined the relationship be- tween mussel length, total mass, and dry tissue mass using analysis of covariance (ANCOVA), with season as the classification variable. Average tissue mass and relative tissue content (dry tissue mass / total dry mass) of mussels in each length class were calculated using predicted (from regressions of biomass on length) total and tissue biomass (Table 1). We used predicted as opposed to raw values to ensure that a true and consistent mean for each length class was achieved (because the mean length of mussels sam- pled from each length class may have varied slightly from sample to sample and was not necessarily the arithmetic mean length of all mussels in that class). Our approach likely introduced very little bias into the analysis, because r 2 values for each regression ex- ceeded 0.94 and intercepts were zero (by definition because the regression was log-log). We then com- pared relative tissue content (arcsine transformed) across length classes and seasons using two-way ANOVA and the a posterJori Tukey's HSD test (Zar 1996). We regressed shell thickness and crushing re- sistance on shell length and compared attachment strength of mussels to files across length classes us- ing ANOVA and Tukey's test. RESULTS Prey-selection experiments.--Common Eiders were size-selective predators on both regular and manipulated tiles during all seasons at both locations (Table 2). Total observed and predicted numbers of mussels consumed in each treatment combination are provided in Ta- ble 3. Preferred length classes differed among treatment groups. At Indian Point, preferences differed among seasons (X 2 = 267.4, df = 9, P < 0.0001) and tile types (regular vs. manipu- lated; spring, X 2 = 13.6, df = 3, P = 0.004; sum- mer, X 2 = 22.5, df = 3, P < 0.0001; winter, X 2 = 74.8, df = 3, P < 0.0001). During spring, sum- mer, and fall ducks generally preferred smaller length classes, especially 19 to 28 mm, and avoided larger classes (Table 2). However, in winter, large mussels were selected (Table 2). During all seasons in which comparisons could be made, ducks tended to take smaller mussels from manipulated tiles than they did from cor- responding regular tiles (Table 2). TABLE 2. Results of simple X 2 analysis of Common Eider prey-selection experiments. "Reps" refers to the number of tiles pooled to obtain the final result. Symbols associated with mussel length classes are as follows: - -, strongly avoided; -, avoided; 0, eaten randomly; +, selected; + +, strongly selected. These classifications are based on cell X 2 values. If the cell X 2 -> 6.63 (P = 0.01 at 1 df), the class was strongly selected or avoided, and if 6.63 -> X 2 -> 3.84 (P = 0.05 at 1 df), the class was selected or avoided. These divisions are not intended as a posteriori tests, but rather as a means of standardizing levels of preference and avoidance. For overall X 2 comparisons, df 3 and critical value of P = 0.05 was at X 2 = 7.815. Values in parentheses indicate percentage of the total X 2 value attributable to that cell; high values indicate strong selection or avoidance. Mussel length class (mm) Site Season Type Reps X 2 10 to 19 19 to 28 28 to 37 37 to 50 Indian Point Spr 1995 Regular 17 19.8 0 (3%) + (30%) - (23%) -- (44%) Indian Point Sum 1995 Regular 23 106.5 ++ (9%) ++ (17%) -- (45%) -- (29%) Indian Point Spr/sum 1995/ Manipulated 12 15.1 + (33%) - (28%) 0 (2%) - (37%) 1996 Indian Point Aut 1995 Regular 9 27.8 0 (5%) + (14%) - (81%) 0 (0%) Indian Point Win 1996 Regular 16 182.5 -- (35%) ++ (7%) ++ (7%) ++ (51%) Indian Point Win 1996 Manipulated 5 16.2 0 (19%) (72%) 0 (0%) 0 (9%) Barr Road Spr/sum 1995 Regular 20 22.4 + + (34%) - (22%) 0 (0%) -- (44%) Bart Road Spr 1996 Regular 9 84.3 0 (4%) + + (65%) -- (15%) -- (16%) Bart Road Spr/sum 1995/ Manipulated 6 29.6 - (34%) ++ (65%) 0 (1%) 0 (0%) 1996 Preferences at Barr Road differed among years (X 2 = 73.7, df = 3, P < 0.0001) and tile types (1995, X 2 = 40.8, df = 3, P < 0.0001; 1996, X 2 = 12.6, df = 3, P = 0.006). Common Eiders feeding on regular tiles preferred mussels in the 10 to 19-mm class in 1995 and the 19 to 28- mm class in 1996 (Table 2). They avoided the smallest mussels on manipulated tiles in both years, strongly preferred length class two, and fed randomly on larger classes (Table 2). Pref- erences for length classes also differed among sites on manipulated tiles (X 2 = 40.6, df = 3, P < 0.0001) and regular tiles in 1995 (spring, X 2 = 15.0, df = 3, P = 0.002; summer, X 2 = 46.2, df = 3, P < 0.0001). At Indian Point, ducks feeding from manipulated tiles during spring and summer preferred the smallest length class, whereas at Barr Road, they selected class- two mussels (Table 2). Ducks eating mussels from regular tiles in 1995 generally preferred prey of 19 to 28 mm (and to a lesser degree, 10 to 19 mm) at Indian Point and 10 to 19 mm at Barr Road (Table 2). Mussel characteristics.--In 1995, mussels from Indian Point tended to be smaller than those at Barr Road, although lengths also varied within locations (Fig. 2). No samples were taken in 1996 at Bart Road, but the size-frequency dis- tribution of mussels from exclosure experi- ments at Indian Point during winter 1995-1996 and spring 1996 suggest that mussel sizes re- mained relatively constant throughout the ex- TABLE 3. Total observed and expected (based on a null hypothesis of random removal) number of mussels of each length class eaten by Common Eiders throughout the experiment. Numbers are corrected for losses from sources other than ducks (see text) and are rounded to the nearest whole digit. 10 to 19mm 19 to 28mm 28 to 37mm 37 to 50mm Site Season Type Obs Exp Obs Exp Obs Exp Obs Exp Indian Point Spr 1995 Regular 277 265 131 106 55 74 21 40 Indian Point Sum 1995 Regular 351 298 182 133 21 86 10 48 Indian Point Spr/sum 1995/1996 Manipulated 64 48 13 23 16 14 3 10 Indian Point Aut 1995 Regular 187 171 99 82 15 48 26 27 Indian Point Win 1996 Regular 109 232 144 107 97 68 91 35 Indian Point Win 1996 Manipulated 90 75 16 37 21 20 21 16 Bart Road Spr/sum 1995 Regular 319 274 98 123 82 82 22 43 Bart Road Spr 1996 Regular 105 123 110 55 13 34 3 19 Barr Road Spr/sum 1995/1996 Manipulated 16 35 32 15 10 8 7 7 a) 700 600 500 400 300 200 100 0 <10 10-19 19-28 28-37 37-50 Size class (mm) IP left [] IP right >50 b) 200 15o 100 5O 0 <10 10-19 19-28 28-37 37-50 >50 Size class (ram) BarrRd rock [] BarrRd weir Fie. 2. Length frequency distributions of mussels collected from (A) Indian Point and (B) Barr Road in summer 1995. Within each site, mussels were col- lected from two areas (Indian Point left and right, Barr Road rock and weir). Different bars represent the different areas at each site. a) 0.3 0.25 0.2 (0.15 'o. 0.05 Jul/95 Aug/95 Dec/95 Mar/96 May/96 Season 10-19 mm 19-28 mm []28-37 mm L137-50 mm b) 0.16 0.14 0.12 0.1 0.08 0.06 0.04 0.02 o Jul/95 Aug/95 De1/2/95 Mar/96 May/96 Season 10-19 mm 19-28 mm E:]28-37 mm Lq37-50 mm FIG. 3. Tissue content of mussels collected during each sampling period. Values for (A) average dry-tis- sue mass of mussels were calculated as an average of predicted (from regression) dry-tissue content for mussels in each length class. (B) Proportion tissue biomass was calculated using values from part (A) and similar ones predicting total dry-mussel bio- mass. Error bars represent _+ 1 SE. periment, with only a small increase in the pro- portion of large mussels in the population in spring 1996 (D. Hamilton unpubl. data). Mussel density was very high at both study sites throughout the experiment. On tiles, den- sity was approximately 3,000 mussels per m 2. In samples collected in summer 1995 (Fig. 2), densities were 3,420 _+ SD of 968 and 1,170 + 263 at two areas of Barr Road, and 7,330 + 2,422 and 2,870 + 1,509 at two areas of Indian Point (excluding mussels 410 mm long). Den- sities declined but still remained high at Indian Point in 1996, with approximately 1,700 mus- sels per m 2 (D. Hamilton unpubl. data). Al- though these ambient densities were highly variable, densities on tiles were within the range of those found naturally. The high den- sities of mussels found throughout the study area suggested that variation in underwater search time would be small (see Discussion). Dry-tissue mass of mussels increased expo- nentially with length (Table 1), and the rela- tionship varied with collection period (AN- COVA, length x date interaction F = 7.5, df = 4 and 240, P 0.0001), although trends ap- peared consistent across seasons (Fig. 3A). Generally, mussels contained the most tissue in July (just before spawning) and the least in March and August (Fig. 3A). However, when predicted length-class means were considered using ANOVA, although the main effect of sea- son was significant (F = 2.64, df = 4 and 197, P = 0.035), the result was due primarily to a difference between July and March (Tukey's a) 1.2 1 0.6 0.4 0.2 Thickness = 0.0625 x length. '65 b) 250 200 150 100 LL 50 10 20 30 40 50 60 Mussel length (mm) Force =2.754 x length .8. . ...: - ..- 10 20 30 40 50 60 Mussel length (mm) FiG. 4. (A) Regression of shell thickness on mus- sel length. (B) Regression of force (in Newtons) re- quired to crush mussels on mussel length. Results of significance tests are provided in text. HSD test). Relative tissue biomass also varied among mussel length classes (Fig. 3B), and the variation differed among seasons (ANOVA, season x length class interaction, F = 68.4, df = 12 and 185, P < 0.0001). During all sample months, the main effect of length class was highly significant (F >-141.6, df = 3 and 37, P < 0.0001 for each sample). All length classes differed from each other; the smallest mussels had the highest relative tissue biomass and the largest mussels the lowest proportion of tissue (Tukey's HSD test). However, differences among length classes were substantially higher in May, July, and August than they were in De- cember and March (Fig. 3B). Both shell thickness (r 2 = 0.75, df = 1 and 41, P < 0.0001; Fig. 4A) and resistance to crushing (r 2 = 0.86, df = 1 and 46, P < 0.0001; Fig. 4B) were positively correlated with shell length. When crushed, shells usually failed first in the middle or toward the rear (away from the a) 1800 1600 o 1200 1000 800 v 600 400 Eo 200 -- 0 Jul/g5 AukS5 DealS5 Mar/g6 May/g6 Season b) 120 100 m 20 0 Jul/g5 AukS5 De5 Mar/g6 May/g6 Season FiG. 5. (A) Estimated shell mass consumed per day by Common Eiders based on average consump- tion of 130 g dry tissue. Values were calculated based on predictions from regression of tissue and shell on mussel length. (B) Ratio of energy gained to work done by Common Eiders eating mussels of each length class during a day. Work is based on cost as- sociated with crushing 130 g dry tissue of mussels of each length class. umbo) of one valve or the other Attachment of mussels to the substrate differed among length classes (ANOVA, F = 140.1, df = 3 and 556, P < 0.0001); the smallest mussels were the easiest to detach and largest ones the most difficult. Costs and benefits.--In any season, Common Eiders that met their energetic requirements by feeding on large mussels would have to con- sume more shell biomass than ducks that fed on small ones (Fig. 5A). However, as described above, differences among length classes in shell mass consumed while eating a set amount of mussel tissue varied among sehsons. In Decem- ber and March, the difference in shell masses consumed by eating equivalent dry-tissue bio- mass of the smallest versus largest length clas- ses of mussels was relatively small (7.8 and 4.9%, respectively) (Fig. 5A). During other months, however, the differences were larger, ranging from 18.5% in May and July to 52.5% in August (Fig. 5A). Energy intake relative to work was always highest when large mussels were taken (Fig. 5B). The ratio of energy gain relative to cost (estimated as the work required to crack a mussel) increased steadily across length classes, and trends appeared similar for all seasons, although the increase in benefit was somewhat lower during May and August than it was at other times of the year (Fig. 5B). DISCUSSION Common Eiders feeding in our experiment were size-selective predators. This has been shown before, although preferred sizes of prey vary among studies (Raffaelli et al. 1990, Nys- trm et al. 1991, Guillemette et al. 1996). How- ever, unlike earlier studies of prey selection in this species, we attempted to quantify and con- trol availability of prey of different lengths. This could be achieved only on tiles; we had no control over the ambient size distribution of prey in the mussel bed, which could strongly influence our results (see below). However, our experiment provided standard prey composi- tion and equal search time for all length classes once a tile was encountered by a foraging Com- mon Eider (because each class covered an equivalent surface area). Thus, our study offers a starting point for assessing size-selective pre- dation. This is important, because prey selec- tion is influenced by availability of both prof- itable (Stephens and Krebs 1986) and unprof- itable prey size classes (Elner and Hughes 1978, Ward 1991). Although differences in mussel size distributions between tiles and the sur- rounding area may have influenced prey selec- tion by Common Eiders, differences in mussel density probably had little effect. Mussels in the experimental area were superabundant, and ambient densities of mussels were similar to those on tiles, so birds never had to look for prey during a dive once they reached the bot- tom. Hence, variation in patch quality related to prey density was eliminated from consider- ation. Relative abundance of different-sized mus- sels on tiles changed somewhat as trials pro- gressed because ducks removed more mussels of some length classes than others. This was unavoidable, but it probably had only a small effect on the results. Tiles were checked daily and removed as soon as noticeable predation had occurred. In most cases, this meant that tiles were removed before any size class was completely eliminated or even substantially re- duced. Hence, foraging ducks still had a rela- tively equal choice of all four classes in the same immediate area. In situations where ducks quickly removed all or most prey of one length class (mostly in winter 1996; see below), the change in prey availability as the trial pro- gressed may have had some effect on selection. However, this effect would be a conservative bias (following a null hypotheses of random prey removal) because elimination of preferred prey would force a switch to the next best choice, hence broadening the diet and reducing the apparent degree of selectivity. Therefore, because we found significant prey selection, this artifact of our experimental design was not a serious problem. Seasonal variation.--Common Eiders selected mussels of different lengths at different times of the year. For most of the year at Indian Point, ducks feeding from regular tiles preferred the two smallest length classes, 10 to 19 mm and 19 to 28 mm, and avoided large mussels. How- ever, in winter this trend was reversed; ducks strongly avoided 10 to 19-mm mussels and se- lected others (Table 2). The largest mussels (37 to 50 mm) were the most preferred, although others were also selected, probably after all large mussels had been removed from tiles. This switch may have been related to changes in costs and benefits of feeding on prey of dif- ferent lengths at different times of the year (see below). If Common Eiders selected prey that maxi- mized short-term energy gain, they should al- ways have fed on large prey. The ratio of en- ergy gained to work done (Fig. 5B) varied little across seasons. Notwithstanding the increase in shell thickness and force required to crush large mussels, energy intake appeared to be maximized by taking the largest prey. This was a consistent trend across seasons, because al- though crushing resistance was measured only once during the experiment, shell mass relative to length (and therefore thickness) varied little through the year. Because we did not attempt to quantify handling time, which probably is higher for large mussels (Draulans 1982, De- Leeuw and VanEerden 1992), the relative ben- efits of consuming large prey may be some- what lower than indicated. However, based on relative tissue biomass (Fig. 3A) and costs as- sociated with crushing mussels (Fig. 5B), Com- mon Eiders would have to consume 15 to 20 small mussels in less time than it takes them to eat one large mussel to reverse the order of en- ergetic profitability of the length classes. Costs associated with detaching prey from the sub- strate may also have slightly reduced the rela- tive benefits of large prey because these re- quired the greatest force to remove. This effect would have been minimal, however, if the force required to remove mussels from the substrate did not alter profitability of different size clas- ses, as was found for Tufted Ducks (Aythyafu- ligula) feeding on zebra mussels (Dreissena po- lymorpha) in Europe (Draulans 1982). Common Eiders have a nail on the upper bill, and a strong grasping action, allowing them to easily remove even large mussels from the substrate (Meire 1993). Therefore, even with the inclusion of handling time and costs of mussel detach- ment, large prey would probably remain the most energetically profitable when considered on an individual mussel basis. The overall benefits of prey to foraging ani- mals depend on more than just energetic prof- itability per prey item. Foraging ducks must si- multaneously consider benefits of different prey and costs, such as those associated with ingestion of shell. The presence of food in the digestive tract limits consumption of other food (Ball 1990, 1994), so if Common Eiders consume a high proportion of shell in a feeding bout, less room will be available for mussel tis- sue. Consumption of less mussel tissue per for- aging bout will probably require more feeding bouts, and certainly more feeding time, to ob- tain the necessary food intake each day. In all seasons, small mussels had the highest propor- tional tissue biomass (more shell per tissue), but variation in tissue content among length classes in winter was lower than during the rest of the year (Fig. 3B). During most of the year, Common Eiders could therefore reduce shell ingestion substantially by feeding on smaller length classes. Notwithstanding the high indi- vidual value of large prey, lower shell content may have made small mussels the most bene- ficial prey for most of the year However, during winter little variation existed among length classes of mussels in the amount of shell con- sumed by ducks while acquiring their daily re- quirement of mussel tissue. Thus, because the added cost of high shell ingestion associated with large mussels was reduced, it may have been most beneficial for Common Eiders to feed on larger prey, which offered more ener- getic benefit on a short-term basis, during win- ter. Our results appear to be consistent with the shell-mass minimization hypothesis of Bustnes and Erikstad (1990). Ducks selected prey that allowed them to minimize shell ingestion when large differences between length classes were evident (most of the year). However, when shell mass was least variable among mussel length classes, Common Eiders appeared to switch tactics in an attempt to maximize short-term energy intake by taking large mussels. This suggests that shell content was an important component of overall prey value. Our results are also consistent with those of Barras et al. (1996), who studied acorn selection by Wood Ducks (Aix sponsa), and of Zwarts and Blomert (1992), who studied preferences of Red Knots (Calidris canutus) for Macoma balthica. In both cases, birds selected prey that minimized shell intake relative to the amount of tissue ingested. However, our results do not provide conclu- sive proof of the shell-minimization hypothe- sis. Even in spring and summer, the difference among mussel length classes in the ratio of en- ergy intake relative to work (Fig. 5A) was con- siderably higher than the difference in shell content (Fig. 5B). As discussed above, if other costs such as handling time and detachment of mussels from the substrate were factored in, larger mussels would likely still be more ener- getically profitable on a per prey basis. By sug- gesting that ducks are attempting to minimize shell ingestion, we assume that total energy in- take is sufficiently limited by shell accumula- tion and passage time to offset any benefit of feeding on more energetically profitable length classes. We have no data to support this as- sumption. Our results indicate that energy maximization appears not be the primary fac- tor influencing prey selection by Common Ei- ders. However, to fully test the shell-minimi- zation hypothesis, researchers should combine an experimental study such as ours with ana- lyses of shell retention time and effects of shell accumulation in the digestive tract on prey in- gestion rates. It would also be useful to consid- er seasonal variation in lipid, protein, and ash content of different length classes of mussels. Geographic variation.--Selection of prey by Common Eiders varied among locations. At In- dian Point in 1995, mussels in the 19 to 28-mm class were the most preferred by ducks, al- though they also selected smaller prey during summer. At Barr Road during the same time, Common Eiders strongly preferred the 10 to 19-mm length class. Tiles had the same length composition at each site, but naturally occur- ring mussels at Indian Point were smaller than those at Barr Road (Fig. 2). This result lends support to the risk-averse foraging hypothesis. In an environment (Barr Road) where mussels were generally large and therefore unprofitable in terms of shell ingestion (but not energy yield), ducks selected smaller prey that mini- mized the risk of inadvertently taking large mussels. Draulans (1984) found that as the proportion of large, unprofitable prey in the population in- creased, ducks took smaller mussels, either to reduce the risk of taking one too large to han- dle (and therefore unprofitable), or because large mussels were more highly variable in profitability, and therefore presented a higher risk (the risk-averse foraging hypothesis). Draulans' suggestion that birds took small prey to avoid those that were too large is supported by our data. Ducks in our study could eat the larger mussels, but these mussels may have been less beneficial during most of the year be- cause of large shell masses. During spring 1996, ducks at Barr Road pre- ferred 19 to 28-mm mussels, contrary to results from the previous year. During 1996, all tiles were placed in the rock area of Barr Road, where mussels were somewhat smaller the pre- vious year (Fig. 2B). The previous year, tiles were mixed evenly among the two areas. Therefore, the experimental area in 1996 prob- ably was not as heavily dominated by very large mussels as it was in 1995; accordingly, ei- ders selected mid-sized prey (the same as at In- dian Point). This result highlights the impor- tance of considering the natural availability of prey in experiments such as ours. Because size- frequency distributions and feeding locations of ducks were relatively consistent among years at Indian Point, comparisons at that site were not biased by changes in the underlying prey-size distribution to which ducks were ac- customed. However, specific Common Eider feeding locations at Barr Road changed from 1995 to 1996, and the natural prey-size distri- bution experienced by ducks in the two years probably differed. Although it is difficult to draw firm conclusions without knowledge of the exact size distribution of prey associated with each area in each year, variation in the am- bient size distribution of mussels may have led to the shift in preferred prey size. In future studies of this type, it would be prudent to avoid study areas such as Barr Road where size distributions of prey in the environ- ment surrounding tiles differed, and birds fo- cused on different sections of the site in differ- ent years (although we had no way of predict- ing this in advance). However, comparing en- vironments with different prey-size distributions (Indian Point vs. Barr Road) is de- sirable because it allows tests of the effects of ambient prey availability. It is also noteworthy that when dealing with highly mobile preda- tors such as Common Eiders, it is virtually im- possible (without radio-tagging and tracking birds) to control their exposure to different prey sizes before the experiment. Hence, if ex- periments are to be done under natural condi- tions (which is important if we want realistic estimates of prey selection), we have to accept that some level of variation in experience, which may influence results, is unavoidable. Variation in prey availability on tiles.--We made a further test of the risk-averse foraging hy- pothesis by comparing regular and manipulat- ed tiles at each location. At Indian Point, ducks feeding on manipulated tiles during spring and summer preferred mussels of 10 to 19 mm, as opposed to 19 to 28 mm preferred by ducks feeding on regular tiles. When the available proportion of less beneficial prey (i.e. high shell content) increased, ducks responded by select- ing smaller mussels, possibly to avoid mistak- enly taking a large one. Surprisingly, selection of mussels also varied with tile type during winter Given that large mussels were preferred on regular tiles during winter, ducks should have continued to select them on manipulated tiles. However, in this sit- uation all length classes were taken randomly, except 19 to 28 mm, which was strongly avoid- ed. Possibly, 10 to 19-mm mussels were taken incidentally because ducks were feeding on large mussels located in the same tile section as small prey. The smallest mussels (which are more mobile than larger individuals) some- times attached on top of large mussels on ma- nipulated tiles. If ducks selected these small prey, they could easily remove them without dislodging the larger mussels, which were at- tached more firmly to the underlying tile. How- ever, if ducks were selecting the larger prey (as they did in winter), small mussels on top of them may have been removed at the same time. It is also possible that the small sample (n = 5) of manipulated tiles in winter provided unclear results due to insufficient replication. It should be noted that the problem of ducks accidentally removing mussels that were at- tached to others was likely only an issue for the manipulated tiles in winter. On regular tiles in all seasons, the four length classes were sepa- rated, and although mussels tended to clump (which we attempted to minimize by separat- ing mussels when tiles were assembled), this would not influence selection because birds would be just as likely to take a clump of one length class as another. In seasons other than winter, Common Eiders selected only small- and medium-sized mussels. As indicated above, on manipulated tiles these ducks would have had no difficulty removing small prey without disturbing the underlying larger ones. The risk-averse foraging hypothesis was not supported when manipulated and regular tiles were compared at Barr Road. Birds actually ap- peared to select somewhat larger mussels from manipulated tiles than they did from regular ones, especially in 1995. We have no explana- tion for this except that again, results may be suspect due to a very small sample at this lo- cation (n = 6), and because we were forced to pool 1995 and 1996 data for manipulated tiles at Barr Road due to small sample sizes (note above that differences existed among years on unmanipulated tiles in this location). Hence, these results should be interpreted with cau- tion. Conclusions.--Common Eiders are size-selec- tive predators, and their preferences vary with season and risk of taking poor-quality prey. We found support for both aspects of optimal prey selection that we studied (i.e. the shell-mass minimization hypothesis and the risk-averse foraging hypothesis), although the latter should probably be investigated further due to questionable results in one area where we had limited replication. We examined only a few of many decisions confronting predators each time they dive. Although we controlled for oth- er factors (e.g. diving depth and prey availabil- ity), we did not address effects of handling time, dive duration, and when during dives birds fed on mussels from our tiles, any of which could have influenced our results (Beau- champ et al. 1992, DeLeeuw and VanEerden 1992, Guillemette et al. 1992). Prey selection under natural conditions is complicated and of- ten is difficult to explain using simple models (Ball 1994). The fact that we observed signifi- cant, interpretable patterns suggests that the factors we examined are important to foraging Common Eiders. Some of the variability in previous estimates of prey selection by Common Eiders may be ex- plained by these findings, and by the fact that previous studies have not incorporated season- al variation in prey value. Raffaelli et al. (1990) found that Common Eiders preferred mussels 10 to 25 mm, and that these were large relative to the available population. They collected Common Eiders in December and January, when, according to our results, large prey should be preferred. Similarly, Guillemette et al. (1996) reported that Common Eiders fed in winter on a modal mussel length of 8 mm, when the modal availability was 3 to 4 mm (al- though they ascribed part of the difference to different collection times for mussels and ducks). Nystr6m et al. (1991) found that Com- mon Eiders selected mussels 17 to 18 mm, but that these were smaller than the average of those available. They attributed this to attempts by the ducks to minimize salt intake by eating small mussels. However, they collected their data in September and October, when, accord- ing to our findings, selection of smaller mussels may also be favored to minimize shell inges- tion. These results highlight the importance of considering factors such as prey availability, lo- cal background conditions, and season in stud- ies of prey selection. Foraging choices by Com- mon Eiders are influenced by season and by the relative abundance of undesirable prey. These ducks are capable of adjusting their feeding patterns relative to seasonal changes in prey quality. This is a regular, repeatable pattern. They apparently are less adept at handling un- predictable variation within seasons in the abundance of prey of different sizes, and they appear to respond to increased relative abun- dance of poor-quality prey by taking smaller mussels. ACKNOWLEDGMENTS We thank Corina Brdar, Cindy Doherty, and Kim Smith for field assistance. We are grateful to Cedric Boone for assistance, and to the University of New Brunswick (Saint John) Faculty of Engineering for al- lowing us to use the tensometer. Dave Ankney, Eliz- abeth Boulding, John Fryxell, Matthew Litvak, Tim- othy Wootton, and Peter Yodzis provided comments on earlier versions of this manuscript. We thank Margaret Peterson, James Sedinger, and two anony- mous reviewers for their helpful comments. Re- search was carried out at Huntsman Marine Science Centre. Funding was provided by grants to DJH from the Institute for Wetlands and Waterfowl Re- search (Bonneycastle Fellowship) and the Delta Wa- terfowl Foundation. Additional funding was provid- ed by a Natural Sciences and Engineering Research Council postgraduate fellowship, Ontario Graduate Scholarship, and Elgin Card Avian Ecology Fellow- ship (University of Guelph) to DJH, and an NSERC Research Grant to TDN. LITERATURE CITED BALL, J.P. 1990. Active diet selection or passive re- flection of changing food availability: The un- derwater foraging behavior of Canvasback ducks. NATO ASI Series 20:95-107. BALL, J.P. 1994. Prey choice of omnivorous Canvas- backs: Imperfectly optimal ducks? Oikos 70: 233-244. BARRAS, S.C., R. M. KAMiNSKI, AND L. A. BRENNAN. 1996. Acorn selection by female Wood Ducks. Journal of Wildlife Management 60:592-602. BEAUCHAMP, G., M. GUILLEMETTE, AND R. YDENBERG. 1992. Prey selection while diving by Common Eiders, Somateria mollissima. Animal Behaviour 44:417-426. BIDARD, J., J. C. THERRIAULT, AND J. BIRUB12. 1980. Assessment of the importance of nutrient recy- cling by seabirds in the St. Lawrence Estuary. Canadian Journal of Fisheries and Aquatic Sci- ences 37:583-588. BUSTNES, J. O., AND K. E. ERIKSTAD. 1990. Size selec- tion of common mussels, Mytilus edulis, by Com- mon Eiders, Somateria mollissima: Energy maxi- mization or shell weight minimization? Cana- dian Journal of Zoology 68:2280-2283. CANTIN, M., J. B12DARD, AND H. MILNE. 1974. The food and feeding of Common Eiders in the St. Lawrence estuary in summer. Canadian Journal of Zoology 52:319-334. DELEEuW, J. J., AND M. R. VANEEERDEN. 1992. Size selection in diving Tufted Ducks Aythyafuligula explained by differential handling of small and large mussels Dreissena polymorpha. Ardea 80: 353-362. DRAULANS, D. 1982. Foraging and size selection of mussels by the Tufted Duck, Aythya fuligula. Journal of Animal Ecology 51:943-956. DRAULANS, D. 1984. Sub-optimal mussel selection by Tufted Ducks Aythya fuligula: Test of a hypoth- esis. Animal Behaviour 32:1192-1196. DRAULANS, D. 1987. Do Tufted Duck and Pochard se- lect between differently sized mussels in a sim- ilar way? Wildfowl 38:49-54. EGERRUP, M., AND M.-L. H. LAURSEN. 1992. Aspects of predation on intertidal blue mussels (Mytilus edulis L.) in the Danish Wadden Sea. Report to the ICES Shellfish Committee. ELNER, R. W., AND R. N. HUGHES. 1978. Energy max- imization in the diet of the shore crab, Carcinus maenas. Journal of Animal Ecology 47:103-116. GU1LLEMETTE, M., A. REED, AND J. H. HIMMELMAN. 1996. Availability and consumption of food by Common Eiders wintering in the Gulf of St. Lawrence: Evidence of prey depletion. Canadian Journal of Zoology 74:32-38. GUILLEMETTE, M., R. C. YDENBERG, AND J. H. H1M- MELMAN. 1992. The role of energy intake rate in prey and habitat selection of Common Eiders So materia mollissima in winter: A risk-sensitive in- terpretation. Journal of Animal Ecology 61:599- 610. HAMILTON, D. J. 1997. Community consequences of habitat use and predation by Common Eiders in the intertidal zone of Passamaquoddy Bay. Ph.D. dissertation, University of Guelph, Guelph, On- tario. HAMILTON, D. J., C. D. ANKNEY, AND R. C. BAILEY. 1994. Predation of zebra mussels by diving ducks: An exclosure study. Ecology 75:521-531. HILGERLOH, G. 1997. Predation by birds on blue mus- sel Mytilus edulis beds of the tidal fiats of Spiek- eroog (southern North Sea). Marine Ecology Progress Series 146:61-72. HUGHES, R. N. 1979. Optimal diets under the energy maximization premise: The effects of recogni- tion time and learning. American Naturalist 113: 209-221. MEIRE, P.M. 1993. The impact of bird predation on marine and estuarine bivalve populations: A se- lective review of patterns and underlying caus- es. Pages 197-243 in Bivalve filter feeders in es- tuarine and coastal ecosystem processes (R. E Dame, Ed.). NATO ASI Series G: Ecological Sci- ences, vol. 33. Springer-Verlag, Berlin. MEIRE, P.M., AND A. ERVYNCK. 1986. Are Oyster- catchers (Haematopus ostralegus) selecting the most profitable mussels (Mytilus edulis)? Animal Behaviour 34:1427-1435. NYSTRC)M, K. G. K., O. PEHRSSON, AND D. BROMAN. 1991. Food of juvenile Common Eiders (Somater- ia mollissima) in areas of high and low salinity. Auk 108:250-256. RAFFAELLI, D., V. FALCY, AND C. GALBRAITH. 1990. Ei- der predation and the dynamics of mussel-bed communities. Pages 157-169 in Trophic relation- ships in the marine environment (M. Barnes and R. N. Gibson, Eds.). Aberdeen University Press, Aberdeen, United Kingdom. STEPHENS, D. W., AND J. R. KREBS. 1986. Foraging the- ory. Princeton University Press, Princeton, New Jersey. VANWINKLE, W. 1970. Effect of environmental fac- tors on byssal thread formation. Marine Biology 7:143-148. WARD, D. 1991. The size selection of clams by Afri- can Black Oystercatchers and Kelp Gulls. Ecol- ogy 72:513-522. ZAR, J.H. 1996. Biostatistical analysis, 3rd ed. Pren- tice-Hall, Upper Saddle River, New Jersey. ZWARTS, L., AND A.-M. BLOMERT. 1992. Why Knot Calidris canutus take medium-sized Macoma bal- thica when six prey species are available. Marine Ecology Progress Series 83:113-128. Associate Editor: 1. S. Sedinger