Group foraging by a stream minnow: shoals or aggregations?

Anim. Behav., 1992.44, 393-403

Group foraging by a stream minnow: shoals or aggregations?

M A R Y C. FREEMAN* & GARY D. GROSSMANSchool of Forest Resources, University of Georgia, Athens, GA 30602, U.S.A.

(Received 30 May 1991; initial acceptance 25 August 1991;final acceptance 30 September 1991; MS. number: A6066)

Abstract. The importance of social attraction in the formation of foraging groups was examined for astream-dwelling cyprinid, the rosyside dace, Clinostomus funduloides. Dace arrivals and departures atnatural foraging sites were monitored and tested for (1) tendency of dace to travel in groups, and (2)dependency of arrival and departure rates on group size. Dace usually entered and departed foraging sitesindependently of each other. Group size usually affected neither arrival rate nor departure probability.Thus, attraction among dace appeared weak; foraging groups most often resulted from dace aggregatingin preferred foraging sites. The strongest evidence of social attraction was during autumn, when dacedeparture probability often decreased with increasing group size, possibly in response to increased threatof predation by a seasonally occurring predator. Dace also rarely avoided conspecifics, except when anaggressive individual defended a foraging site. Otherwise, there was little evidence of exploitative compe-tition among dace for drifting prey or of foraging benefits in groups, because group size usually did notaffect individual feeding rates. These results suggest that the benefits of group foraging demonstratedunder laboratory conditions in other studies may not always apply to field conditions.

Many animals forage in groups, and social foragingmay provide a variety of benefits to group members,including higher or less variable foraging rates andlower risks of predation (Pulliam & Caraco 1984;Pitcher 1986). The frequent and widespreadoccurrence of group foraging by stream fishes(Reimers 1968; Seghers 1974; Mendelson 1975)suggests that individual fish benefit from feeding ingroups, and that groups form as a result of socialattraction among individuals. In support of thishypothesis, several laboratory studies have demon-strated potential benefits of group foraging forfreshwater fish (as discussed below). Fieldstudies, however, have focused on habitat use andspecies segregation within mixed-species groupsvMendelson 1975; Gorman 1988), rather than onmechanisms promoting group foraging. Conse-quently, we have little information on the extent towhich group foraging by stream fish in situ is asocial behaviour. We attempted to quantify theimportance of social attraction (i.e. attractionamong individuals) on group foraging by a streamminnow under field conditions.

Recent studies suggest that freshwater fish maybenefit from foraging in social groups (termed*Present address: Institute of Ecology, University ofGeorgia, Athens, GA 30602, U.S.A.

'shoals', Pitcher 1983) through some of the same. mechanisms demonstrated for flocking birds

(Pulliam & Millikan 1982). For example, in labora-tory tests in which food is hidden in discretepatches, individual fish can locate food morequickly when part of a group than when alone(Pitcher et al. 1982; Magnan & FitzGerald 1984;Street & Hart 1985). Shoaling also may decreaseindividual risk of predation through severalmechanisms, including early predator detection,predator confusion, and evasive manoeuvres by thegroup (reviewed by Godin 1986 and Pitcher 1986).As a result of anti-predator advantages, shoalingfish may allocate more time to foraging and less topredator vigilance (field test: Seghers 1981; lab-oratory tests: Magurran & Pitcher 1983; Magurranet al. 1985; Morgan & Colgan 1987).

Although laboratory studies clearly demonstratethat stream fish may benefit from feeding in groups,these studies do not address the possible influenceof habitat heterogeneity on group formation.Stream fish often occupy structurally diverse habi-tats composed of a heterogeneous mix of sub-strate types, current velocities and water depths.Numerous studies (e.g. Grossman & Freeman1987 and references therein) have demonstratedthat stream fish use the available habitat in-a

0003-3472/92/090393+ 11 $08.00/0 © 1992 The Association for the Study of Animal Behaviour393

394 Animal Behaviour, 44,3

Table I. Predicted patterns for regressions of arrival and departure rates ( K ) as functions of group size (A")

Mechanism of group formation(effect of group sizeon foraging rate)*

Arrival rate versus group sizet

Open forager pool Closed forager poolDeparture rate versus

group sizet

Aggregation (no benefits or costs ofgroup foraging)

Social attraction (benefits of groupforaging exceed costs)

Aggregation (costs of groupforaging exceed benefits)

No relationshipY=aPositive, linearY=a+bX

Negative, linearY=a-bX

Negative, linearY=a(lf-X)Positive, curvilinearY=(a+bX)(N-X)

Negative, curvilinearY=(a-bX)(N-X)

Positive, linearY=aXSignificant negativequadratic termY=(a-bX)XSignificant positivequadratic term

*For range of group sizes observed in the field.tin arrival rate regressions a represents site attractiveness, b represents the effect of each group member on arrival rate(Caraco 1979a, 1980), and Nis the number of dace available to feed in a given site when the forager pool is closed. Indeparture rate regressions a is the per capita probability of departure, and b represents the effect of each group memberon departure probability.

non-random manner. Consequently, suitable for-aging habitat may be patchily distributed, and fishmay aggregate in the best feeding sites even thoughthey do not specifically benefit from group foraging.In fact, foragers may interfere with each other andcontinue to forage in groups as long as the benefitsof feeding in a particular location outweigh thecosts of competition (Pulliam & Caraco 1984).Thus, habitat heterogeneity may cause stream fishto aggregate (i.e. form groups through individualsindependently selecting the same habitat: Williams1964; Breder 1967; Caraco 1979a) in the absence ofsocial attraction.

We investigated group foraging behaviour of aHorth American stream minnow, the rosysidedace, Clinostomus funduloides. Dace commonlyforage alone and in groups in our study sites, wherethey are the most abundant water column fishpresent. Dace primarily forage on drifting inverte-brate prey in deeper areas with low to moderatecurrent velocity (Grossman & Freeman 1987), and

. habitat selection by dace appears to be an energymaximizing decision^ (Hill 1989}. However, theeffects of social attraction'oil habitat use by thisminnow are unknown; groups of dace may rep-resent aggregations in favourable habitat patches,or dace may choose sites on the basis of both habi-tat suitability and ihe presence of conspecifics. It$so is possible that dace actually compete withconspecifics when aggregations form in suitableforaging sites.

To distinguish between social attraction andaggregation as mechanisms promoting group for-aging, we observed dace in natural foraging sitesand asked three questions.

(1) Do dace move among foraging sites alone orin groups? A possible consequence of social attrac-tion would be for dace to move among foragingsites in schools or pairs (Partridge 1980). In thiscase, dace should enter and depart at a givenforaging site in groups. In contrast, if groups resultfrom individuals aggregating in foraging sites,then individuals should primarily forage indepen-dently of conspecifics, and single dace should fre-quently arrive to and depart from a given foragingsite.

(2) Do arrival and departure rates at a foragingsite, expressed as functions of group size, provideevidence of attraction (or avoidance) among for-agers? If foraging groups result from social attrac-tion, as opposed to aggregation, then this should bereflected in the relationship between arrival and/ordeparture rate and group size (Table I). We followedthe approach used by Caraco (1979a, 1980) to modeldynamics of bird flocks to derive alternative predic-tions of arrival rate relationships to group size. Ifdace groups are aggregations (and net foragingbenefits are independent of group size) then the rateof dace arrivals to a particular site should be a, aconstant value reflecting the attractiveness of thesite. If dace are attracted to (or avoid) conspecifics,then the arrival rate to a site should be a+bX,

Freeman & Grossman: Group foraging by minnows 395

where X is group size and b reflects the increase(or decrease) in site attractiveness for each fishadded to the group. Avoidance among foragerswould imply that groups are aggregations andthat net foraging benefits decrease with groupsize. These equations assume there is an openpool of available foragers (Caraco 1979a, 1980).The number of dace available to feed at a particu-lar site (AO may in fact be constant, so that arrivalrates also depend on how many individuals arenot already at the site (i.e. N—X, Table I; Caraco1980).

Departure rate of dace from a foraging siteshould increase as a linear function of group size ifthere is neither attraction nor avoidance amongdace foraging together at a site. If, however, there isan added effect of attraction or avoidance amongdace, then the relationship of departure rate togroup size should include either a negative orpositive quadratic term (Table I).

Arrival or departure rates plotted across the fullpossible range of group size may actually be dis-continuous or non-linear if, for example, foragersonly benefit from group foraging, or only interferewith each other, after groups reach a threshold size.In the field, however, group sizes on the negative(less beneficial) side of a threshold should be rareunless habitat or foragers were limiting. Thus,regressions of arrival and departure rates versusgroup size at a particular site should reflectrelationships among foragers for those particularconditions and the range of group sizes observed inthe field.

(3) Do dace feeding or aggression rates vary as afunction of group size? This was not a test for socialattraction per se, but rather a question of whetherwe could identify costs or benefits of group foragingunder field conditions. We did not witness any pre-datory attacks on foraging dace, and so we couldnot quantify group-size effects on predation risk. Ifforagers in groups devote less time to vigilance,however, this should result in a positive feedingrate-group size relationship. Conversely, a lack ofrelationship between feeding rate and group sizewould indicate that dace do not derive foragingbenefits in groups. Finally, a negative group sizeeffect on feeding rate would indicate a cost to groupforaging because of increased exploitative compe-tition. Similarly, if per capita aggression increaseswith group size, then greater interference amongdace could be a cost of group foraging. Examininggroup-size effects on feeding and aggression rates

also allowed us to test whether variation in dacearrival or departure dynamics in groups corre-sponded to variability in the net benefits of groupforaging.

METHODS

Study Site and Field Methods

Our research site was located at the U. S. D. A.Forest Service, Coweeta Hydrologic Laboratory(Macon County, North Carolina), in the southernAppalachian Mountains. We worked in CoweetaCreek, a fifth-order stream, and one of itsfourth-order tributaries, Ball Creek. Stream seg-ments used in this study flow through relativelyundisturbed, mixed-hardwood forest. Riparianvegetation is well developed and includes thickclumps of Rhododendron maxima and Kalmialatifolia. A more detailed description of thestudy site is provided by Grossman & Freeman(1987).

These mountain streams provide good habitat forobserving water column fish, because interspersedriffles and pools create natural impediments toforager movement. Individual cyprinids normallyforage within 10-15-m-long pools and may remainin particular pools for months (Hill & Grossman1987). Also, heterogeneous current velocities andsubstrates (ranging from silt and sand to boulders)may subdivide suitable habitat for water columnfish within pools. Thus, it is possible to observeforagers in patches of suitable habitat that are smalland relatively discrete. We worked in four studypools, situated 40-190 m apart, that ranged from 9to 16 m in length, with widths of 3-6 m. The par-ticular observation site used in a given pool varied,depending on flow conditions and where groups ofdace were foraging. We conducted observationsfrom August through to November in 1988 and1989, and during April and early May 1989.Because dace spawn during spring or early summer(Davis 1972), reproductive activity should not haveaffected our observations.

During each observation, we recorded activity ina 40 x 40 cm quadrat in an area where one or moredace were foraging. This quadrat size ensured thatindividual dace within the quadrat generally wereno more than 5-7 body-lengths apart. Data fromother investigators suggest that 5-10 body-lengthsis the maximum distance between interacting fish(Seghers 1981; Pitcher et al. 1983; Helfman 1984).

396 Animal Behaviour, 44, 3

Thus, dace within the quadrat probably were closeenough together to interact. This quadrat size wastoo small, however, to contain large groups thatformed in some habitats. Therefore, we placedquadrats in relatively discrete foraging locations;i.e. areas that individuals moved in and out of, butwhere fish usually did not feed in the immediatelysurrounding area. We used coloured markersplaced on the stream bottom to delineate a quadratat a selected foraging site.

We observed foraging activity by snorkelling; fishdid not appear disturbed by a stationary observer.We used a tape-recorder (enclosed in a waterproofhousing, with a waterproofed microphone attachedto a snorkel) to record the numbers and species offish present and fish movements in and out of thequadrat for periods of 25-40 min. During mostobservations we also recorded displacements orchases among foragers. A displacement involvedone fish swimming laterally towards a target indi-vidual, which either shifted position or movedlaterally back towards the aggressor. .Chasesinvolved an aggressor either swimming directlytowards another fish, or approaching and pursuinga fish from behind. Finally, during most obser-vations we counted feeding strikes either by all indi-viduals or by focal individuals. We counted allfeeding strikes made by the foraging group if thegroup was small and feeding activity was relativelylow. Otherwise, we haphazardly selected focal indi-viduals from the group for 1-2-min observations,attempting to watch as many different individualsduring sessions as possible. We could not usuallysee the particles that foragers captured from thedrift, except when fish struck at debris which theysubsequently spat out. During the 1989 obser-vations, we tabulated rejected prey (typicallystrikes at debris) separately from other strikes.

Data Analysis

We transcribed tape-recorded data by tabulatingactivity (i.e. initial group size, arrivals, departures,aggressive acts, feeding strikes) occurring in thequadrat during successive 5-s intervals. To addressour first question, whether dace moved among for-aging sites in groups, we compared the frequenciesof arrivals and departures by zero, one, two, threeand four or more dace (in 5-s intervals) withexpected frequencies generated from Poisson distri-butions (i.e. assuming independent movement). Weused chi-squared goodness-of-fit tests (alpha = 0-025,

to control type I error at 0-05 for two non-independent tests), and combined categories whennecessary so that no expected frequencies were lessthan 1.

To test for either attraction or repulsion of dace togroups of conspecifics, we regressed average arrivaland departure rates (per 5-s interval) on group size(= number of dace present at the beginning of a 5-sinterval). Regressions were calculated separately forobservations from diflferent sites and dates. Arrivalsand departures were approximately Poisson distri-buted; many 5-s intervals had no movement in or outof the quadrat, and variances among 5-s intervalswere approximately equal to average rates. Wetherefore used weighted least-squares regression tocorrect for heteroscedasticity (Neter et al. 1989);weights were the inverses of arrival (or depar-ture) rates predicted by unweighted least-squaresregressions. We only included average rates inregressions for those group sizes with enough 5-sinterval observations so that either (1) the standarderror was 30% of the estimated rate or less, or (2)when the estimated rate was 0-2. or less, the stan-dard error was 0-05 or less, assuming rates followedPoisson distributions. This procedure was necess-ary to avoid biasing regression analyses, by includ-ing poorly estimated average rates (i.e. those forgroup sizes that were infrequently observed). Theonly exceptions were for four data sets in which thelargest group size had frequent arrivals or depar-tures and at least 1 min of observation; these grouprates were included in analyses. We excluded fromregression analyses those 5-s observations duringwhich other water-column fishes were present in theforaging site, except (1) when heterospecifics werepresent for less than 2-5% or more than 90% of the.total observation time, in which cases we pooledobservations with and without heterospecifics for agiven dace group size, and (2) when it was necessaryto pool data to obtain sufficient sample sizesfor analysis for one date when rainbow trout,Oncorhynchus mykiss, were present for 65% of theobservation time.

To test whether dace arrivals (from a closed for-ager pool) or departures were dependent on groupsize, it was necessary to test fits of both linear andquadratic models (see Table I). Quadratic modelsof arrival rates were fit to deviation of group sizefrom mean group size to reduce correlationbetween the independent variables (i.e. x and x2,where x=group size; Neter et al. 1989). Weregressed departure rates through the origin, and

Freeman & Grossman: Group foraging by minnows 397

independent variables were not transformed. Termswere retained in the models if partial F-tests weresignificant at the 0-1 probability level. We also used a0-1 probability level to test overall regression signifi-cance. When both linear and quadratic models weresignificant, we judged the quadratic model to providea better fit if the residuals showed less evidence ofsystematic variation than with the linear model.

We used weighted least-squares analysis to testfor relationships between feeding and aggressionrates and group size, analysing data for each obser-vation set separately. Feeding rate data for focaldace were pooled to estimate (1) average per capitafeeding rate, including rejected particles, and (2) theproportion of bites that resulted in rejected par-ticles, for each observed group size. We estimatedaverage feeding rates only for group sizes with atleast 1 min of observation. Minimum sample sizecriteria for estimating aggression rates for a par-ticular group size were the same as for estimatingarrival and departure rates.

RESULTS

Did Dace Move to and from Observation SitesIndependently, or in Groups?

Dace usually entered and departed observationsites independently of each other. Foragers did notenter or leave in groups more often than expectedunder the hypothesis of independent movement in59% of the observation sets. However, groups oftwo or three fish entered sites significantly(P< 0-025) more frequently than predicted fromPoisson expectations in 25% (14 of 56) of the testedobservation sets. Groups of two or three fishdeparted significantly more frequently thanexpected in 20% (11 of 55) of the tested observationsets. Single dace arrived or departed significantlymore frequently than expected in one observationset each. Non-independence of arrivals during anobservation was not significantly associated withnon-independence of departures (G-test for inde-pendence, G = 0-4532, df= \, P > 0-1). This suggeststhat different motivational stimuli affected dacearrivals and departures.

Did Arrival and Departure Rates at a Foraging SiteDepend on Group Size?

Arrival rates of dace to foraging sites were notrelated (P>0-1) to group size in 60% of all obser-vations (N= 52 sets), indicating that the presence of

Table II. Effects of group size on dace arrival to anddeparture from foraging sites. Results are frequencies ofregression patterns for arrival rate and departure rateversus number of dace present in the foraging site

Regression pattern

Number ofobservation

sets(%)

Arrival rate versus dace numbeoNon-significant (i.e. arrivals 31 (60%)*

independent of group size)Negative (i.e. possible avoidance 15(29%)

among foragers)Positive (i.e. attraction among 6 (12%)

foragers) :Departure rate versus dace number

Linear (i.e. departures 24 (48%)independent of group size)

Positive quadratic (i.e. avoidance 19 (38%)among foragers)

Negative quadratic (i.e. attraction 7 (14%)among foragers)

* Ranges of probability levels associated with non-significant regressions were: P>0-l-0-2, Ar=13 obser-vation sets; P>0-2-0-4, #=9 observation sets; />>0-4,Af=9 observation sets.

conspecifics often did not attract or repulseapproaching foragers (Table II). Probabilitylevels associated with most of the non-significantregression tests exceeded 0-2 (Table II). Note alsothat although we used a 0-1 probability level forregression significance, actual type I error prob-abilities in these tests may have been as high as 0-2because arrival and departure rate regressions werenot independent. Thus, a failure to find a significantrelationship was a conservative conclusion.

Arrival rates increased with group size, and thusprovided evidence of attraction among dace, in only12% of all observations. Arrival rates decreasedwith group size in the remaining 29% of the obser-vation sets. Of the 52 observation sets with sufficientdata for arrival rate regressions (Table II), 36 setshad a wide enough range of group sizes present forus to test the fit of quadratic models. Curvilinearregressions provided the best fits in five of the sixdata sets with positive functions and in two of the15 data sets with negative relationships.

Regressions of arrival rates are difficult to inter-pret without knowing whether or not there wereenough dace in the study areas to maintain arrivalrates across the range of group sizes observed inforaging sites (i.e. whether the forager pool was

398 Animal Behaviour, 44, 3

April/May August September October NovemberFigure 1. Monthly differences in departure rate-group size relationships. The number of observation sets best fit bylinear (•), positive quadratic (D), and negative quadratic functions (0) are plotted for each month. Among monthdifferences were significant (G-test for independence, G= 19-023, df=&, P< 0-025).

open or closed). For example, negative relation-ships between arrival rate and group size (i.e. 29%of the sets), may be interpreted as avoidance amongforagers only if an open pool of dace was avail-able. Interpretation of departure rate-group sizerelationships is less ambiguous than that of arrivalrates because the pool of foragers available to leaveis the number of dace present in the quadrat.

Departure rate regressions suggested that theeffect of group size on dace departure probabilityvaried among observations, from no effect to eitherpositive or negative effects. The only observationsets with non-significant regressions (P> 0-1, five of55 observation sets) had data only for one and twodace present. Otherwise, departure rate alwaysincreased with dace number, and was best describedby a linear function for 24 observation sets (TableII). Thus, for almost half of the observation setsthere was no evidence that the probability of a par-ticular dace leaving a site changed with the numberof conspecifics present, suggesting that groups wereaggregations in which group size did not affect netforaging benefits. The remaining observation setswere best described by quadratic regressions. Daceappeared to avoid larger aggregations during 19observations (i.e. sets with positive quadraticcoefficients). Conversely, there was evidence ofattraction among foragers during seven obser-vations, when dace were less likely to depart largergroups (i.e. sets with negative quadraticcoefficients). ^

The relative frequencies of linear, positivequadratic and negative quadratic departure rate

functions differed significantly among months (Fig.1). Dace generally departed foraging sites inde-pendently of group size during" spring (April andMay) and autumn (October and November),whereas foragers avoided conspecifics in 72% of theobservations made during August and September.In contrast, dace were attracted to larger groups in43% of the November observations. This patternwas similar between years, except that we observedavoidance in October 1988 (when flows werereduced because of a drought from 1985 to 1988)but not in October 1989. This seasonal shift inbehaviour did not correspond to any dramaticchange in size distributions of foraging dace; mostindividuals averaged 5-7 cm total length during allobservations.

In summary, analyses of arrival and departurerates suggest that groups most often were aggre-gations in which dace did not interact. However,there was evidence of avoidance among aggregat-ing foragers on some dates (primarily during latesummer), while there was evidence of attractionamong dace in less than 15% of the observations(primarily during autumn).

Did Group Size Affect Dace Foraging orAggression Rates?

Group size effects on foraging rates

Feeding rate generally was not linearly related togroup size, indicating that dace usually did notcompete with conspecifics or obtain foraging ben-efits in groups. Only six of 31 regressions of feeding

Freeman & Grossman: Group foraging by minnows 399

Table III. Summary of results of linear regression analysestesting for effects of group size (i.e. number of dacepresent in a foraging site) on feeding and aggression rates

Regression analysis

Number ofobservation

sets (%)

Per capita feeding rate versusgroup size

Non-significant 25 (81%)*Negative slope (P<0-.05) 6 (19%)

% feeding strikes resulting inrejections versus group size

Non-significant 9(100%)tPer capita aggression rate versus

group sizeNon-significant 9(100%)t

*Probability levels associated with non-significantregressions were: P>0-I-0-2, N=l observation sets;P>0-2-0-4, N=10 observation sets; P>0-4, #=8observation sets.

tProbability levels associated with non-significantregressions were: P>0-1-0-2, /V=3 observation sets;?>0-2-0-4, N=3 observation sets; P>0-4, N=3observation sets.

rate versus group size were significant (Table III); inall six, per capita strike rates were a decreasingfunction of group size. In addition, regressions ofthe proportion (arcsine transformed) of strikes thatresulted in particle rejection versus group size fornine data sets all were non-significant (Table III).Thus, the general lack of relationship between feed-ing rate and group size was not caused by foragersin larger groups striking more frequently at inedibledebris.

In five of the six cases where group size had anegative effect on feeding rate (Table III), dace wereless likely to remain in larger aggregations (i.e. dacedeparture probability increased with group size).However, during 10 other observations in whichdeparture probability increased with group size, percapita feeding rate was unrelated to the number ofdace present. Thus, intraspecific exploitative com-petition could not account for most observations ofdace avoiding larger groups.

Group size effects on aggression among daceThere was no evidence that the probability of

being chased or involved in aggression changed pre-dictably with the number of conspecifics present,and thus variation in this relationship did not corre-spond to whether or not dace were attracted to

groups. Regressions of per capita aggression rateversus group size for nine observation sets all werenon-significant (Table III). Aggression by dace wasinfrequent overall, averaging 0-016 chases and dis-placements per individual per 5-s interval (N=60,SD = 0-032), or about one incident per dace every5-2 min. During 13 of 60 observations, however, asingle dace initiated all or nearly all aggressive acts.These aggressors tended to remain in the obser-vation sites throughout observation periods, pri-marily leaving to chase out a conspecific, to meet aconspecific swimming toward the quadrat, or tointercept a drifting particle. On average, aggressionrates were an order of magnitude higher when-asingle aggressor was present than when several daceinitiated chases or displacements (means=0-0581versus 0-0057 incidents/5 s in sites with versus with-out a single aggressor; Wilcoxon test, z = 5-13,P< 0-0001). Nine of the 13 observations with asingle aggressive dace were during August andSeptember, and observation sets with aggressivedace were characterized by higher water tempera-tures than those without aggressors (means =15-3°C versus 12-6°C, Wilcoxon test, z = 2-59,P<0-01).

Although the average probability of beingchased was unrelated to group size, the presenceof a single aggressive dace apparently affected therelationship between departure probability andgroup size. In eight of the 13 observations witha single aggressive dace, departure probabilityincreased with group size (i.e. there was a positivequadratic relationship between departure rate anddace number). Of the other five observations witha single aggressor, two had insufficient ranges ofgroup sizes to regress departure rates, whereas inthree, all from October or November, departurerate increased linearly with dace number. Dacewere more likely to leave a site in groups of two orthree when an aggressor was present: seven of the12 data sets with non-independent departures wereobservations including a single aggressor (G-testfor association, G=8-581, d f = l , P<0-005). Dacedid not, however, enter sites in groups more fre-quently when a single aggressor was present (G =0-115, df= 1, />>0-1). This at least partly explainsthe lack of association between non-independenceof arrivals and departures (described above).

In summary, the presence of a single aggressivedace could account for most observations ofincreased departure probability as a functionof group size. However, we found no instances of

400 Animal Behaviour, 44,3

feeding benefits, or lower per capita aggression inlarger groups.

DISCUSSIONOur results suggest that dace frequently foragedindependently of each other and that groups mostoften resulted from fish aggregating in profitableforaging locations rather than from social attrac-tion. Although there may have been a seasonal shiftin this behaviour, dace arrival rate rarely increasedas a function of group size, and departure rate mostoften was independent of group size. In addition,individuals usually did not enter or depart forag-ing sites in groups. Assuming dace forage so as tomaximize individual fitness, our results suggest thatdace did not obtain substantial benefits fromforaging near conspecifics.

Group foraging has been shown to benefit fresh-water fishes in laboratory studies. Why would thesebenefits not apply to minnows foraging in fieldconditions? Individuals may benefit from socialforaging via two main mechanisms: (1) enhancedforaging success and (2) decreased risk of pre-dation. Feeding data provided little evidence thatforaging in groups benefits individual dace. More-over, it is unlikely that local enhancement, throughwhich group members benefit by sharing fooddiscovered by others (Pulliam & Millikan 1982),generally is significant for drift feeding fishes.Local enhancement is effective when resources arepatchily distributed, food in patches is shared bygroup members, and groups find or feed in patchesmore efficiently than solitary foragers (Pulliam &Millikan 1982). In laboratory tests, for example,shoaling minnows benefit because food patches arefound more quickly by several searchers, and oncediscovered, food patches can be shared by groupmembers (Pitcher et al. 1982; Magnan & FitzGerald1984)."; Drift feeders use two identifiable resources:suitable feeding sites, and relatively large driftinginvertebrates. Feeding sites may be patchily dis-tributed but presumably are of known distributionto local foragers. Large drifting prey probablyoccur independently through time, assuming drift-ing invertebrates enter the current (and settle) inde-pendently. Thus, a dace usually would not benefitby joining a conspecific that has just taken a bite.

Group foraging could allow drift feeders tospend more time foraging because group membersspend less time in predator vigilance. For example,in laboratory experiments, members of largergroups of cyprinids began foraging more quickly

(Morgan & Colgan 1987), spent less time in coveror in evasive manoeuvres (Magurran & Pitcher1983), and continued to feed longer in the presenceof a model predator (Magurran et al. 1985; but seeRehnberg & Smith 1988 for contradictory results).Similarly, Hill (1989) found that rosyside dace in anartificial stream would not forage unless at leastfour individuals were present in the stream together(in the absence of predators), suggesting that dacewere less timid in larger groups. Our field data,however, provided no evidence of this effect; daceforaged alone (as well as in groups) during allmonthly observations, and solitary foragers hadthe highest strike rates recorded.

Predation pressure on dace at Coweeta may betoo low during much of the year for individuals tobenefit appreciably from anti-predation functionsof shoaling. The known piscivores at the study sitesare northern watersnakes, Nerodia sipedon, beltedkingfishers, Megaceryle alcyon, raccoons, Procyonlotor, and from late autumn through early spring,rockbass, Ambloplites rupestris. Watersnakes areuncommon, and we have only seen evidence of asingle incident of fish predation by raccoons in 8years of research at Coweeta Creek. Kingfisherswere more common in 1988 and 1989 than inseveral years previous (personal observation).Group foraging may not offer much protection,however, against an aerial predator that attacksbefore it can be spotted by any group member(Goodey & Liley 1985; Pitcher 1986), a likelysituation in streams with considerable surfaceturbulence.

The increased tendency during autumn for daceto remain in large groups may have been a responseto a perceived increase in predation risk. Forexample, cyprinids threatened by a predator underexperimental conditions form larger or morecohesive groups (Pitcher et al. 1983; Morgan 1988).Rockbass usually move upstream into the vicinityof the study areas in late autumn and leave by earlyspring (personal observation; Stouder 1990). Wehave no direct evidence, however, that rockbassactually pose a threat to cyprinids in the study sites(rockbass captured from these sites in previousyears usually had empty stomachs; Stouder 1990).

Competition and Aggression Among DaceThe observation of increased departure prob-

abilities with increasing group size during 19observations indicated costs of foraging near con-specifics, especially during late summer. Negative

i

Freeman & Grossman: Group foraging by minnows 401

effects of group size on foraging rate, however,only were apparent during five of these obser-vations. This was surprising given that drift rates oflarge prey (i.e. animals > 2 mm total length, theminimum-sized prey normally consumed by dace,Hill 1989; Stouder 1990) are low at Coweeta, aver-aging less than two prey per min in a 20 x 20 cmsection of water column at dace foraging sites(Freeman 1990). Dace may not compete exploit-atively in groups because foragers are unable todiscriminate prey from inedible debris withoutstriking (as is suggested by observations of dacestriking large drifting debris and frequent rejec-tions). In this case, opportunities for feeding strikeswould often not be depleted even in large foraginggroups.

Avoidance among foragers was most stronglyassociated with the presence of aggressive indi-viduals, and it is interesting to consider what dacemay gain through aggressive behaviour given thatstrike rate generally does not decline with increas-ing group size. We hypothesize that an aggressive(or solitary) individual may have greater access tolarge drifting prey. It is common for several for-agers in a group to strike, in succession, at a largeparticle of drifting debris; aggressive individualsmay often have the first opportunity to strike largeprey. The potential importance of large prey is evi-denced by energetic considerations. Standard meta-bolic requirements for a 6-cm dace (standard length;weight approximately 4 g) in autumn are approxi-mately 91 1 J/day; assuming standard metabolic rateequals 0-7 mg O2/g/h (Facey & Grossman 1990),and 13.6 J/mg O, (Elliott & Davison 1975). A dacewould need to consume at least 43 mg of prey eachday to meet resting metabolic costs (assuming100% assimilation of inverteb'rates'averagmg21 J/mg dry weight), or greater than 800 individual 2-mm long prey (Smock 1980). Because insect massincreases exponentially with body size (Smock1980), foraging dace may require captures ofrelatively large prey to meet their energeticrequirements.

These calculations suggest that because of thelow density of large drifting prey, dace at Coweetamay be limited by the time available for foraging.In this case, the increase in fitness produced byaggressive behaviour may depend on the energeticrequirements of individuals. For example, aggres-sion among flocking birds may increase with tem-perature because reduced energetic requirementsallow more time for aggression (Caraco 1979b,

1980). Standard metabolic rates of dace acclima-tized in the field actually were lowest in summer(15°C) and highest in autumn (10°C), possibly as aresult of gonadal development in autumn (Facey &Grossman 1990). Dominance behaviour by dacemay therefore have been more frequent during latesummer because reduced energetic demandsallowed more time for aggression than duringautumn. Additionally, aggression by three indi-viduals in late October and November 1989 did notresult in increased departure probabilities fromlarger groups. Thus, aggression may become a lesseffective strategy (e.g. for defending access to largeprey) if dace form larger groups during autumn inresponse to some other factor.

Conclusions

We conclude that dace foraged in groups mostoften as a result of individuals aggregating in profit-able foraging locations rather than from attractionamong individuals. Our observations are similar tothose of house sparrows, Passer domesticus, feedingon patches of seed under conditions of low pre-dation risk, in which arrivals were unrelated toflock size at a seed patch (Barnard 1980a, b). Percapita aggression among sparrows increased withflock size, increasing the probability of departure.Per capita aggression was unrelated to group sizein dace; however, intense aggression by particulardace may have increased departure probabilitiesfor targeted individuals. When aggression amongdace was very low, departure probability often wasunrelated to group size. Similarly, Krebs (1974)observed that herons foraging in flocks did notinterfere with each other or directly benefit from thepresence of conspecifics, and departure ratesincreased linearly with flock size.

Our results suggest that benefits of group forag-ing demonstrated for fish in laboratory conditions(Pitcher et al. 1982; Magurran & Pitcher 1983;Magurran et al. 1985; Street & Hart 1985; Morgan& Colgan 1987) may not always apply to groupsforaging under field conditions. However, daceresponses to groups of conspecifics varied consider-ably among our observations. This variation wasonly in part seasonal (e.g. attraction among dacewas more evident during autumn), and a separateanalysis failed to find any correspondence betweendeparture rate-group size relationships and preydrift rate, water depth, or current velocity at theforaging sites (Freeman 1990). Variability in dace

402 Animal Behaviour, 44.3

behaviour may have represented stronger shoalingtendencies by a few individuals, similar to thevariation observed by Helfman (1984) among indi-vidual yellow perch, Perca flavescens, and byMagurran (1986) among individual minnows,Phoxinus phoxinus, responding to a stalking pred-ator. A possible source of variability among dace isoccasional immigration from downstream popu-lations, where dace coexist with rockbass through-out the year and anti-predator behaviour may bemore strongly developed (Seghers 1974; Magurran1986, 1990). Additional field studies of cyprinidforaging behaviour, especially in habitat wherepiscivores are more common, could elucidate theconditions under which social attraction is import-ant to these fishes and the extent to which dacebehaviour in pur study was a result of low predationpressure.

ACKNOWLEDGMENTS

We gratefully acknowledge assistance from thestaff of the Coweeta Hydrologic Laboratory,especially M. L. Rollins, K. Reynolds, B.Cunningham and W. T. Swank. B. J. Freeman,D. J. Stouder and M. G. Flood provided advice andfield help, and D. Nickerson helped with regressionanalysis. G. S. Helfman, H. R. Pulliam, M. J. VanDen Avyle and J. B. Wallace provided valuablecriticism. This project was supported by U. S. D. A.Forest Service Mclntire-Stennis program (GEO-0035-MS to G. D. Grossman), and by the NationalScience Foundation (BSR-8514328 to D. A.Crossley Jr and J. L. Meyer et al.). The School ofForest Resources and the Graduate School of theUniversity of Georgia provided generous financialsupport to M.C.F.

REFERENCES

Barnard, C. J. 1980a. Flock feeding and time budgets inthe house sparrow (Passer domesticus L.). Anim.Behav., 28,295-309.

Barnard, C. J. 1980b. Equilibrium flock size and fac-tors affecting arrival and departure in feeding housesparrows. Anim. Behav., 28,503-511.

Breder, C. M., Jr. 1967. On the survival value of fishschools. Zoologica, 52,25-40.

Caraco, T. 1979a. Ecological response of animal group sizefrequencies. ln:$Statistical Distributions in EcologicalWork (Ed. by J. K. Ord, G. P. Patil & C. Taille), pp.371-386. Fairland, Maryland: International Coopera-tive Publishing House.

Caraco, T. 1979b. Time budgeting and group size: a test oftheory. Ecology, 60,618-627.

Caraco, T. 1980. Stochastic dynamics of avian foragingflocks. Am. Nat., 115,262-275.

Davis, R. M. 1972. Age, growth, and fecundity of therosyside dace, Clinostomus funduloides Girard.Chesapeake Sci., 13,63-66.

Elliott, J. M. & Davison, W. 1975. Energy equivalents ofoxygen consumption in animal energetics. Oecologia(Berl.), 19,195-201.

Facey, D. E. & Grossman, G. D. 1990. The metabolic costof maintaining position for four North American streamfishes: effects of season and velocity. Physiol. Zoo!., 63,757-776.

Freeman, M. C. 1990. Foraging behavior of the rosysidedace, Clinosiomus funduloides: the importance of socialinteractions. Ph.D. thesis, University of Georgia,Athens.

Godin, J.-G. J. 1986. Antipredator function of shoaling inteleost fishes: a selective review. Naturaliste Can., 113,241-250.

Goodey, W. & Liley, N. R. 1985. Grouping fails toinfluence the escape behaviour of the guppy (Poeciliareticulata). Anim. Behav., 33,1032-1034. •

Gorman, O. T. 1988. The dynamics of habitat use in aguild of Ozark minnows. Ecol. Monogr., 58,1-18.

Grossman, G. D. & Freeman, M. C. 1987. Microhabitatuse in a stream fish assemblage. /. Zoo/., Land., 212,151-176.

Helfman, G. S. 1984. School fidelity in fishes: the yellowperch pattern. Anim. Behav., 32,663-672.

Hill, J. 1989. The energetic significance of microhabitatuse in two stream fishes. Ph.D. thesis, University ofGeorgia, Athens.

Hill, J. & Grossman, G. D. 1987. Home range estimatesfor three North American stream fishes. Copeia, 1987,376-380.

Krebs, J. R. 1974. Colonial nesting and social feeding asstrategies for exploiting food resources in the great blueheron (Ardea herodias). Behaviour, 51,99-134.

Magnan, P. & FitzGerald, G. J. 1984. Mechanisms respon-sible for the niche shift of brook charr, Salvelinusfontinalis Mitchill, when living sympatrically with creekchub, Semotilus atromaculatus Mitchill. Can. J. Zoo/.,62,1548-1555.

Magurran, A. E. 1986. Predator inspection behaviour inminnow shoals: differences between populations andindividuals. Behav. Ecol. Sociobiol., 19,267-273.

Magurran, A. E. 1990. The inheritance and developmentof minnow anti-predator behaviour. Anim. Behav., 39,834-842.

Magurran, A. E., Oulton, W. J. & Pitcher, T. J. 1985.Vigilant behaviour and shoal size in minnows. 2.Tierpsychol.,67,l61-m.

Magurran, A. E. & Pitcher, T. J. 1983. Foraging, timidityand shoal size in minnows and goldfish. Behav. Ecol.Sociobiol., 12,147-152.

Mendelson, J. 1975. Feeding relationships among speciesof Notropis (Pisces: Cyprinidae) in a Wisconsin stream.Ecol. Monogr., 45, 199-230.

Morgan, M. J. 1988. The effect of hunger, shoal size andthe presence of a predator on shoal cohesiveness in

Freeman & Grossman: Group foraging by minnows 403

bluntnose minnows, Pimephales notatus Rafinesque. J.FishBioi, 32,963-971.

Morgan, M. J. & Colgan, P. W. 1987. The effects of pred-ator presence and shoal size on foraging in bluntnoseminnows, Pimephales notatus. Environ. Biol. Fish., 20,105-111.

Neter, J., Wasserman, W. & Kutner, M. H. 1989. AppliedLinear Regression Models, 2nd edn. Boston: Irwin.

Partridge, B. L. 1980. The effect of school size on thestructure and dynamics of minnow schools. Anim.Behav., 28,68-77.

Pitcher, T. J. 1983. Heuristic definitions of fish shoalingbehaviour. Anim. Behav., 31,611-613.

Pitcher, T. J. 1986. Functions of shoaling behaviour inteleosts. In: The Behaviour ofTeleost Fishes (Ed. by T. J.Pitcher), pp. 294-337. Baltimore: The Johns HopkinsUniversity Press.

Pitcher, T. J., Magurran, A. E. & Winfield, I. J. 1982. Fishin larger shoals find food faster. Behav. Ecol. Sociobiol.,10, 149-151.

Pitcher, T. J., Magurran, A. E. & Allan, J. R. 1983. Shiftsin behaviour with shoal size in cyprinids. Proc. Br.Freshw. Fish. Con/, 3,220-228.

Pulliam, H. R. & Caraco, T. 1984. Living in groups: isthere an optimal group size? In: Behavioural Ecology.An Evolutionary Approach. 2nd edn (Ed. by J. R.Krebs & N. B. Davies), pp. 122-147. Oxford: BlackwellScientific Publications.

Pulliam, H. R. & Millikan, G. C. 1982. Social organiz-ation in the nonreproductive season. In: A vian Biology.Vol. K/(Ed. by D. S. Farner, J. R. King &K..C. Parks),pp. 169-197. London: Academic Press.

Rehnberg, B. G. & Smith, R. J. F. 1988. The influence ofalarm substance and shoal size on the behaviour ofzebra danios, Brackydanio rerio (Cyprinidae). /. FishBiol., 33,155-163.

Reimers, P. E. 1968. Social behaviour among juvenile fallChinook salmon. /. Fish. Res^BdCan., 25,2005-2008.

Seghers, B. H. 1974. Schooling behaviour in the guppy(Poecilia reticulata): an evolutionary response topredation. Evolution, 28,486-489.

Seghers, B. H. 1981. Facultative schooling behaviour in thespottail shiner (Notropis hudsonius): possible costs andbenefits. Environ. Biol. Fish., 6,21-24.

Smock, L. A. 1980. Relationships between body size andbiomass and aquatic insects. Freshw. Biol., 10,375-383.

Stouder, D. J. 1990. Dietary fluctuations in stream fishesand the effects of benthic species interactions. Ph.D.thesis, University of Georgia, Athens.

Street, N. E. & Hart, P. J. B. 1985. Group size and patch. location by the stoneloach, Noemacheilus barbatulus,

a non-visually foraging predator. J. Fish Biol., 27,785-792.

Williams, G. C. 1964. Measurement of consociationamong fishes and comments on the evolution of school-ing. Publ. Mus. Mich. St. Univ. Biol. Ser., 2,349-384.