Nordic Society Oikos

Eight Reasons Why Optimal Foraging Theory Is a Complete Waste of Time

Author(s): G. J. Pierce and J. G. Ollason
Source: Oikos, Vol. 49, Fasc. 1 (May, 1987), pp. 111-117
Published by: Wiley on behalf of Nordic Society Oikos
Stable URL: http://www.jstor.org/stable/3565560 .
Accessed: 19/10/2014 19:51

Your use of the JSTOR archive indicates your acceptance of the Terms & Conditions of Use, available at .
http://www.jstor.org/page/info/about/policies/terms.jsp

.
JSTOR is a not-for-profit service that helps scholars, researchers, and students discover, use, and build upon a wide range of
content in a trusted digital archive. We use information technology and tools to increase productivity and facilitate new forms
of scholarship. For more information about JSTOR, please contact support@jstor.org.

Wiley and Nordic Society Oikos are collaborating with JSTOR to digitize, preserve and extend access to Oikos.

http://www.jstor.org

This content downloaded from 73.188.188.250 on Sun, 19 Oct 2014 19:51:38 PM

All use subject to JSTOR Terms and Conditions
 FORUM is intended tofacilitate
communication between readerandauthorandreaderand
FORUM reader.Comments, viewpoints
welcome.Discussion
orsuggestions
aboutimportant
causedbyspeculative
issuesinecology,
controversial
papersare
orterminology
e.g. theory mayalso
FORUM be included.Contributions
topicand summarizing
shouldbe as conciseas possible.A summary
theargument maybe included.Reference
the
introducing
shouldbe madeonlyto
workbasicto thetopic.A lighter
prose,designed to attract willbe permitted.
readers, For-
FORUM malresearchreports,albeitshort,willnotbe accepted.
Ed.

Eightreasonswhyoptimal is a complete
theory
foraging wasteoftime

G. J. Pierceand J. G. Ollason, CultertyField Station,Univ.ofAberdeen,Newburgh,Ellon, Aberdeenshire

AB4
OAA, Scotland(no significanceis to be attachedto theorderof theauthors'names)

Summary. Wepresent oftheapplication

a seriesofcriticisms of timalwithrespectto criteriathatmay be evaluated in-
optimizationtheoryto thebehaviour and morphology ofani- dependentlyof a knowledgeof thefitnessoftheanimals
mals,usingtheexampleofoptimalforaging theory. Thecrit- et
icismsareindependentandpresented indecreasing orderofim- (see, e.g., Pyke al. 1977). Users of optimalforaging
portance.We arguethatoptimization theory is inappropriate theory tryto formulate optimaldecisions,withrespect
theproducts
forinvestigating ofevolution,thatanimalsshould to the independentcriteria,to generatetestablequali-
notbe expectedto be optimal,thatit is notpossibleto test tative and quantitativepredictionsabout foragingbe-
whether theyareoptimal. Wefurther suggestthatitis notpos- haviour.
sibletotestwhetherbehaviour hasbeenselected tofulfil
specific
thatsuchtestshavenotbeencarriedout,andthatno
functions, The criticismsare presentedin decreasingorder of
optimizationmodelofforaging behaviourhasbeensupported. importance.In each case, a criticismis exploredon the
Appealsto theheuristicvalueofthetheory are inappropriateassumptionthat the more fundamentalobjections can
becausetheyencourage interpretations
unjustified ofthebehav- be
iourofanimals. rejected. If one of the morefundamentalarguments
is accepted,thenthelessercriticisms merelyreferto the
propertiesof false premises.
Our intentis to provoke discussion.Too often,crit-
icismsof optimalforagingtheoryare dismissedas "well
Introduction
known",and quietlysweptunderthe carpetwithoutre-
In thispaper we presenta seriesof criticisms of optimal futation.If theyare refutabletheyshould be refuted,
foragingtheoryand of the way testsof the theoryhave not merelydenied. If theycannotbe refuted,the work
been conducted.Some of thesecriticisms have received criticizedshould be abandoned forthe waste of timeit
considerableattentionin the literature,othersare rela- is.
tivelynew. Theyreferto flawsat all levelsofthetheory:
its epistemologicalbasis, theoreticalvalidity,and em-
1. What does naturalselectionmaximize?
piricalvalue. While we do not question the validityof
the theoryof evolution,it will be apparentthatthe ar- By definition,reproductivefitnessis maximizedby nat-
gumentswe presenthave widerimplicationsforthe va- ural selection(e.g. Williams1970): but what does this
lidityof adaptationistexplanationsof thebehaviourand mean?
morphologyof animals. Optimal foragingtheoryrequiresthatthe reproduc-
The basic tenetof optimalforagingtheoryis this:for- tiveoutputof an animalcan be explainedin termsofthe
aging behaviourhas been shaped by naturalselection, rewardsit achieves(food, a mate,surviving by avoiding
so that foragingstrategieswhichmaximizefitnesswill predation)throughperformance of theactivitiesnecess-
existin nature,and theseforagingstrategieswillbe op- aryforsurviving and reproducing,thatperformance can
be explained in termsof phenotypicattributessuch as
strengthand speed, and thatthese attributesare heri-
Thecontents ofthispaperandthereplyon p. 118corresponds table.
withtheaimsoftheForum.On accountoftheirgeneralinter-
esttheEditorhasacceptedthepapersinspiteoftheirexcess- Because selection acts upon individual animals,
ivelength. whereas genes are perpetuatedas units, the way in

OIKOS 49:1 (1987) 111

This content downloaded from 73.188.188.250 on Sun, 19 Oct 2014 19:51:38 PM

All use subject to JSTOR Terms and Conditions
Fig.1 (above).Thesetofpossiblegenotypes maybe regarded as a g-dimensionalspace,eachdimension beinga locus.A particu-
largenotype maybe defined as a point(g,,g2,..., gg)inthisspace.Thispointmapsinsomewayto a particular phenotype, which
likewisecanbe regardedas a point(q, CP2,... .pn)inann-dimensional spaceeachdimension ofwhichcorresponds toa phenotypic
variable.The animal'sbehaviour is dividedup intomseparateactivities thejthofwhichyieldsa reward, r1j,whichis a function,
f,j,ofthephenotypic
variables.Themdistinct rewards defineanm-dimensional spaceandtheparticularsetofrewards (rll,r12 ....
r,m) mapsto thereproductive output,r2.f2is themapping function.
The phenomenological relationships thatareobservable areshowninFig. 1 (below).Thegenotypic properties maybe inferred
onlyempiricallyfromthephenotypic properties. The relationshipbetweenphenotype andreproductiveoutputis also inferred
The argument
empirically. developedintheAppendixsuggests thatr,j,flj,andf2arereifications
thataredevoidofexplanatory
content.

whichgenes influencereproductiveoutputis crucialto If rewardsachieved fromdifferent activitiesare not in-

thevalidityof optimalforagingtheory.Fig. 1 illustrates
possible mappingsof genotypeon to reproductiveout-
put, throughphenotypeand rewardsobtainedfromac-
tivities.These relationshipsare set out moreformally in
the Appendix.
For performance in activities(e.g. therateoffeeding)
to be maximizedbynaturalselection,thefollowingcon-
ditionsmusthold:
(1) Each activitymustbe objectivelydefinable.
(2) Performancein differentactivitiesmust depend
upon non-intersecting setsofphenotypicattributes.
(3) The underlying phenotypiccharactersmustbe heri-
table.
(4) The contribution to reproductiveoutputbyeach ac-
tivitymust be a monotonicincreasingfunctionof
the rewardobtained in thatactivity.
(5) The contributions to reproductiveoutputby differ-
ent activitiesmust be independent,i.e. increasing
the rewardfromone activitymustnot lead to a re-
ductionin the rewardobtainedfromanotheractiv-
ity.
(6) The rewardfromeach activityis maximizedby a
unique set of values forthe controllingset of phe-
notypiccharacteristics.
dependent,thereis no reason to expectperformance in
anyone activityto be maximized.If performance in two
activitiesis determinedby overlappingsets of pheno-
typicattributes,performancein bothcan be maximized
only if changesin those attributesaffecting both activ-
itiesalterbothrewardsin thesame direction.If twoani-
mals withdifferent phenotypescan both achieve maxi-
mal rewardsin an activity,one cannot know a priori
which(if either)alternativewill occur.
Now, considerwhat is knownabout foraging.Opti-
mal foragingtheorypredictsthatthe rate of energyin-
take willbe maximizedonlyifforagingis an independ-
entactivity.However,it is sometimesobservedthatthe
need to avoid predatorsmightconstrainforagersto feed
at less than the maximalrate. This means eitherthat
theseactivitiesare not independentand thereis no rea-
son to expectperformancein eitheractivityconsidered
on itsown to be maximized,or thattheactivityhas been
misidentified, the independentactivityobserved actu-
ally being foragingwhile avoidingpredators.This im-
plies thatit is impossibleto identifyactivitiesa priori,
and it is thereforeimpossibleto obtainevidencethatin-
dependentactivitiesexist. Either it is assumed thatin-
dependentactivitiesexist and it is possible to identify
themby lookingforrewardswhichare maximized,or it

112 OIKOS 49:1 (1987)

This content downloaded from 73.188.188.250 on Sun, 19 Oct 2014 19:51:38 PM

All use subject to JSTOR Terms and Conditions
mustbe assumedthatitis possibleto identify activities, vince themselvesthat the organ, howeverit was con-
whichmaythenbe examinedto discoveriftheyarein- stituted, was well designedto fulfilits function. Se-
dependent.In bothcases thereis no escapefromcir- lectionhas no foresight and can actonlyuponexisting
cularargument, and it mustbe concludedthat"activ- structures:the variantsperpetuatedare those which
ities"are merelyartifacts of thewaybiologists look at currently contribute positivelyto reproductive output.
animals. As the eye was evolving,itsstructurewas changingand
Ifwhatanimalsdo cannotbe dividedintoindepend- the way in which it contributedto fitnessmust have
entactivities then,in Lewontin's(1978a) words,"We been changing.It is not meaningfulto regardevolution
areleftin thehopelesspositionofseeingthewholeor- as havingsolved a problembecause what an observer
ganismas adaptedtothewholeenvironment". The ani- mightperceiveas the problemwas constantlychanging
malsthatleave themostoffspring are reproductively(Ollason in press a).
thefittest,butto statethatnaturalselection maximizes The argumentapplies irrespectiveof the relativebal-
fitnessin thissenseis hardly Optimalfor- ance betweentherateofchangeof theenvironment
illuminating. and
agingtheory assumes that naturalselectionwill maxi- the rate of evolution. Selective pressures are deter-
mizetherewards obtainedbyanimalsengaging ininde- minedbothby the environment and by the animals
pendentactivities, and thatby doingso reproductiveavailable forselection.Even in a constantenvironment,
outputis maximized. Sinceitis impossibleto definethe selectivepressureswill constantlychange as, and be-
rewards ortheactivities theonlything cause, the animalsevolve. The differencebetweenhu-
exceptcircularly,
naturalselectioncan be assumedto maximize is repro- man designand evolutionis not just thata humande-
ductiveoutput. signerknowsin advance the problemto be solved and
the materials available to solve it: evolution cannot
solve problemsbecause theveryprocessof evolutionary
2. Animalsare not designed
change constantlyredefinesthe materialavailable to
Optima, by definition,are the best solutionsto prob- workwithand the problemto be solved.
lems,undergivenboundaryconditions.For theconcept Consequently,functioncan be nothingmore than
of optimalityto be applicable to behaviour,it mustbe tautologousredescription of the consequencesof struc-
possible to view behaviouras solvingproblems.Since ture.As Nagel (1961) has observed,structureand func-
animals have evolved, these solutions must have tionevolve simultaneously, and are inseparable.
evolved: evolutionmust,in some sense,solve problems. Inevitablyanimalspossess whatmightbe regardedas
The problemwhichis solved by a particularpiece of beautifullyconstructedorgans and behaviour, which
behaviour,or by an organ, may be expressed by as- contributeto theirsurvivaland reproduction.However,
cribinga functionto thatbehaviouror organ.Optimiza- whatwe mightregardas the currentfunctionof behav-
tion theory requires that the functionof behaviour iour or structurescannot be assumed to tell us very
should mean somethingmore than a tautologousre- much about the functionsof its evolutionaryantece-
descriptionof its consequences. The observed behav- dents, which must have been different.Optimization
iourmustrepresenttheculmination(or currentstate)of models assume that the functionhas always been the
the evolutionaryprocessof solvinga problem. same and thattheorganor behaviourchangedto corre-
Most biologists acknowledge that evolution is not spondmorecloselyto theoptimum,and as suchmisrep-
purposeful:it does not designorganismsin thewaythat resentthe natureof the evolutionaryprocess.
a Divine Creator mightdesignthem. Further,the fact
that an animal would die if a particularorgan was re-
3. Optimalstrategiesmay not occur in nature.
moved does not implythatthe organ would inevitably
have evolved (Williams1966). Neverthelessit has been Evenifnatural
selection
didtendtogiverisetooprimal
consideredusefulto look at evolutionas thoughit was structureand behaviour,thereare severalreasonswhy
solvingproblems,althoughorganismsare the products we mightnot expect to findoptimalanimals:
of naturalselectionon randommutationand recombi- (1) Optimalstrategiesmay not have evolved yet,or,
nation, the effectresembles objects of design (Ruse as Cody (1974) puts it, populationsmay spend more
1977). To avoid the appearance of imputinga tele- timetrackingmovingfitnessoptima(climbingadaptive
ological component to the evolutionaryprocess, Pit- peaks) than theydo sittingat the summitoptima. Fit-
tendrigh(1958) used the term"teleonomy". ness optima are inevitablymovingas the environment
The analogywithdesignis not a good one: consider and the gene pool change.
the evolutionof the vertebrateeye: At each stagein its (2) If foragershave to learn about the environment
evolutionaryhistory,the organ that evolved into the in order to forageoptimally,the optimalstrategymay
vertebrateeye musthave made a positivecontribution neverbe attained.Ollason (1980), and Macnamara and
to the fitnessof the animalsof whichit was a part. At Houston (1985) addressthequestionof how and animal
each of thesestagesthiscontribution to fitnesscould mightlearnto achievetheoptimumdefinedbythemar-
have been expressedby ascribinga functionto whatthe ginal value theorem(Charnov 1976), and show thatif
organ did, and biologistswould have been able to con- animalslearn as theysuggest,it would take an infinite

8 OIKOS49:1 (1987) 113

This content downloaded from 73.188.188.250 on Sun, 19 Oct 2014 19:51:38 PM

All use subject to JSTOR Terms and Conditions
amountof timeforbehaviourto convergeupon the op-
timum.Perhaps animals can learn optimally:but the
wayan animallearnsimplicitly definesthewayin which
it perceivesits environment,and consequentlydefines
an optimal strategywithrespectto that perceptionof
the environment.In otherwords,all learningstrategies
could be said to be optimal, each in its own terms.
There is no objective criterionwithwhichto compare
learningstrategies:an animal can only be expected to
optimizewithrespectto those featuresof the environ-
ment which are importantto it, but these cannot be
evaluated independentlyof the learningprocess.
(3) The natureof available geneticvariationmaymil-
itate againstthe evolutionof optimalstrategies:there
maynotbe geneticvariationin thephenotypicdirection
postulated,the variationmaybe of thewrongsort(e.g.
the optimal phenotypemay be heterozygous),or the
optimalstrategyof one animal may depend on thatof
others- and ifthe optimalstrategyis a mixtureof phe-
notypesit may not be geneticallyfeasible (Lewontin
1978b).
4. The existenceof optimalstrategiesis untestable
Two methods of testingfor the existence of optimal
strategieshave been proposed in the literature:
(1) The comparativemethod: the behaviouror struc-
turewhichis best adapted is thatwhose formcorre-
sponds most closely to the formtypicalof the be-
haviouror structureassociatedwiththe activitiesin
question (Thompson 1981).
(2) The modellingmethod: what we findin natureis
comparedwith"whatis predicteda priorion theba-
sis of modelsdesignedto mimicthenaturalsystem"
(Cody 1974).
Both approaches assume the abilityof the investigator
to identify,a priori,the rewardan animalseeks whenit
engages in a particularactivity(i.e. the functionof the
behaviour),the relevantcharacteristics of the environ-
ment in whichthe rewardis sought,and the possible
rangeof strategiesavailable to the animal.
Since it is impossibleto knowthe functionof behav-
iour a priori,if the observedbehaviourappears not to
be optimal,it maysimplybe thatthefunctionof thebe-
haviourwas misidentified. Even if observedbehaviour
appears to be optimal,it is possible thatthe behaviour
reallyhas a different function,to whichit is not opti-
mallyadapted. It is alwayspossibleto derive,retrospec-
tively,a functionwithrespectto whichobservedbehav-
iour is optimal,and manyotherfunctionswithrespect
to whichit is not optimal.
Maynard-Smith(1978) acknowledgedthis problem,
and observedthatoptimalitymustbe assumed and that
whatcan be testedis whetherbehaviourfulfilsspecific
functions.Althoughmost studentsof foragingbehav-
iour admitthatthe assumptionof optimalitycannotbe
114
This content downloaded from 73.188.188.250 on Sun, 19 Oct 2014 19:51:38 PM
All use subject to JSTOR Terms and Conditions
tested, it seems to be forgottenthat this means that
therecan be no evidence foroptimalforaging.
5. Functionalhypothesesare untestable
The usual approach in studiesof foragingbehaviouris
to assumethatit is optimaland attemptto findout what
it is optimizedto do. The mostfrequently encountered
functionalhypothesisabout foragingbehaviouris thatit
has been selectedto maximizethe rateof energyintake
whileforaging.This hypothesiscan be testedonlybyus-
ingit to constructa model of whatthe animalis doing.
In such a model it is necessaryto definethe range of
strategiesavailable to the foragerand the environment
in whichit forages.
Models do not attemptto replicatenature exactly,
rathertheyattemptto captureits essence. In optimiza-
tionmodelsofforagingbehaviour,therepresentation of
the environmentis necessarilyan abstraction,in which
realityis simplifiedand thusdistorted.For example,the
marginalvalue theorem(Charnov 1976) assumes that
food occursin discretepatcheswhichbelong to distinct
types,whereasin realitymostpatchesof food probably
have indistinctboundariesand patch qualitymay be a
continuousvariable. Assumptionsmust also be made
about therangeofpossiblebehaviour,and theseare un-
likelyto be accuratewithoutdetailed knowledgeof the
behaviourof the species in question.
For it to be possibleto testthe functionalhypotheses
underlyingoptimizationmodels of foragingbehaviour,
it mustbe possible to provideindependentverification
of the assumptionsmade about the range of strategies
available to foragersand the featuresof the environ-
mentwhichare importantto foragers.If these assump-
tions cannot be verified,confirmationof predictions
mustbe regardedas fortuitousand devoid of explana-
torypower (Ollason in press b).
The featuresof the environment whichare important
to a foragercannotbe determinedindependently of ob-
servingitsbehaviour.It willalwaysbe possibleto iden-
tifya set ofenvironmental withrespectto
characteristics
whichobservedbehaviouris consistentwitha particular
functionalhypothesis,but this process is entirelycir-
cular. By assertingthat animals perceive the environ-
mentin a particularway it would be possible to show
that observed foragingbehaviour was consistentwith
any functionalhypothesis.
6. Optimalforagingmodelshave not been tested
A numberof different problemscan be identifiedre-
gardingthevalidityoftestsofoptimalforagingmodels:
(1) Some published"tests"of optimalforagingmod-
els report experimentsconducted under conditions
whichviolatedassumptionsof themodel tested,e.g. us-
to testmodels
ing foragerswhichsearch systematically
assumingrandomsearch(Krebs et al. 1983,Pyke 1984).
Many existingmodels of foragingbehaviour assume
OIKOS 49:1 (1987)
thatforagersoptimizea singlebehaviouralparameter, differencebetweenthe data and the predictions.What
all othersbeing constant(e.g. Pulliam 1974, Charnov he should thenhave done was to measure the costs of
1976, Oaten 1977). For example, it is usuallyassumed travellingand searchingexperimentally, but insteadhe
thatforagerssearchforprey,and handleprey,withcon- acceptedthe fitobtainedby incorporating the extraun-
stantefficiency.However, there is abundantevidence verifiedassumptionintothe model, and concludedthat
thatpredatorsof all kindscan varythe rateswithwhich the greattitsforagedoptimally.
theysearchforand handleprey:notonlycan theylearn Withoutconductingfurthertests it is impossibleto
to foragemoreefficiently, but theycan foragemoreef- tellwhetherforagersshow partialpreferences,contrary
ficentlywhen hungryor when prey are less abundant to the predictionsof simpleoptimizationmodelsof diet
(e.g. Sih 1982). Pyke (1984) stressedthe need to use selection,because oferrorsin discrimination, long-term
models appropriate to the study animal (and vice learning,inherentvariation,runsof bad luck, simulta-
versa). neous encounterswithprey,or failureof the animal to
Models may be good predictorsof behaviour even be fullyadapted to its conditionsto life (Krebs and
though they contain incorrectassumptions,but this McCleery1984). It is also impossibleto tellwhetherone
means eitherthatthe erroneousassumptionsare irrel- of theseexplanationsis corrector whetherforagersare
evant to the workingof the model, or thattheireffects not "trying"to maximizetheirrate of energyintakein
are counterbalancedbytheinclusionof othererroneous any way at all.
assumptions. It is importantto distinguishbetween Krebs and McCleery(1984) concludetheirdiscussion
thesetwo explanations,since the latterone rendersen- of optimal foragingtheoryby assertingthat the very
tirelyspuriousany supportobtained fora model. simpleoptimizationmodels of foragingbehaviourper-
(2) Most "tests" of optimal foragingmodels seek formremarkablywell, given theirsimplicity.This im-
agreementwithhypotheses,whereaseverybodyknows plies supportfor the underlyingfunctionalhypotheses
thathypothesescan onlybe disproved(Platt 1964). The whichsimplydoes not exist.
problemwithseekingagreementis thatpoorerdata are Currently,thesehypothesesmustbe regardedas hav-
more likelyto supporta model. Many "fits"to predic- ingbeen provisionallydisproved.We suggestthatthere
tionsmightdisappear if more data were available. is not yet any evidence in favourof any optimization
(3) In some studies,the predictionstestedwere not model of foragingbehaviour. Progressin science pro-
unique to the models underconsideration(Krebs et al. ceeds fromthe recognitionof the importanceof dis-
1983, Pyke 1984). crepanciesbetweenpredictionand observation,and the
(4) Often agreementwiththe originalhypothesisis searchfornew theoriesthatreducethesediscrepancies,
obtained only by incorporatingamendmentsto the as- not frombuildingendlessqualifyingclauses on to mod-
sumptionsof themodel. Even ifpredictionsof themod- els to protectthemfromdisproof.
ifiedmodels are upheld, the underlyingfunctionalex-
planationmaythenaccountfora verysmallproportion 8. The heuristicvalue of optimizationmodels
of the observed variabilityin behaviour,and is quite
likelyto be wrong.This is analogous to the 16thCen- Optimalforagingtheoryhas undoubtedlyled to thecol-
turyview of the solar system,in whichthe planetswere lectionof a vast amountof data about foragingbehav-
believed to followcircularorbitsaround the earth. By iour, and while acknowledgingits theoreticaldeficien-
addingepicyclesto the circularmodel planetarymove- cies, variousauthorshave made appeal to the heuristic
mentswere predictedvery accurately,but the under- value of the theory.Thus Oster and Wilson(1978) rec-
lyingcircularmodel was wrongforall that. ommendedthat "the prudentcourse is to regardopti-
malitymodelsas provisionalguidesto further empirical
researchand not necessarilythe key to deeper laws of
7. Optimalforagingmodelshave not been upheld
nature",and Marriset al. (1986) suggestedthat"opti-
No singlepublishedtest of an optimalforagingmodel malitytheoryprovidesusefulguidelinesforthestudyof
that we have encounteredhas provided unequivocal foragingbehaviour,but is not a vehicleforthe precise
supportforthe model. It is totallyirrelevantthatsome simulationor predictionof such behaviour."
predictionsare upheld. Once a modelhas been falsified, This view accords withKuhn's (1970) descriptionof
premise scienceas a puzzle-solvingenterprise,in whichtheories
it is quite incorrectto assertthattheunderlying
was true but thatsome of the otherassumptionsmust are supercedednot because theyhave been falsifiedbut
have been wrong.This is entirelypossibleofcourse,but because the new theoriesare betterpuzzle-solvers.Yet
mustbe confirmedby testingalternativemodels. Thus as Thompson (1981) has pointed out, Kuhn's account
Cowie (1977) found that his great tits (Parus major) was descriptiveand shouldnotbe takenas prescriptive:
stayed longer in each patch than predicted. He ex- "scientistsshould stillattemptto produceunambiguous
plained this by takinginto account the differencebe- predictionsand empirical data to confirmor falsify
tweenthecost of searchingand travelling, whichhe had them".
previouslyassumed to be negligible,and was able to Appeals to heuristicvalue are simplyan excuse for
modifythe predictionsso thattherewas no significant failure.Howevermanydata are generated,referenceto

8* OIKOS 49:1 (1987) 115

This content downloaded from 73.188.188.250 on Sun, 19 Oct 2014 19:51:38 PM

All use subject to JSTOR Terms and Conditions
optimal foragingtheoryleads to spurious interpreta- (2) first-order physicalfitness,flj,beingthe mapping
tionswhichcan onlydetractfromthe understanding of fromphenotypeon to the reward obtained fromen-
foragingbehaviour. gagingin the jth activity.
(3) second-orderphysicalfitness,f2,being the map-
pingfromthe set of rewardsobtainedfromengagingin
Epilogue m activitieson to reproductiveoutput.
Optimizationtheoryhas no place in currentevolution- The rewardobtained by the ith animal engagingin
ary thought:its use is a throwbackto the comfortable the jth activitymay be representedas rlijwhere
determinismof Divine Creation; to the endeavoursof
natural philosophersseeking to demonstratethe wis- rlij = flj (Pi qP2i, *-. )ni),
dom of the Creator.
J. B. S. Haldane (1963) observedthatthereare four The reproductiveoutputof theithanimalmaybe repre-
stagesin thenormalprocessof acceptanceof a scientific sented by the following:
idea:
r2, = f2(rli, r2, ..., rim),
(1) thisis worthlessnonsense;
(2) but perverse,pointof view;
thisis an interesting, Consequently:
(3) thisis true,but quite unimportant;
(4) I alwayssaid so. r2 = f2 (fll (q(Pi, (P2i . , Pni), f12 (qPli, P2i, ..., (Pni). *-
flm ((Pli, (f2i. **. (Pni))
Criticismsof optimalforagingtheoryhave met withall
fourresponses,butcurrentliteraturesuggestsverylittle For it to be possible to justifythe use of optimization
change in the way ecologiststhinkabout foragingbe- theoryin ecology,the mappingsf,j,j = 1,2, ..., m, and f2
haviour.So, whichexcuse is it to be thistime? have to possess the followingproperties:
(1) The m activitiesmusteach be objectivelydefin-
Acknowledgements - We thankG. lason,H. Crick,and M.
able and be independent.
Eastforhelpfuldiscussionsandcomments on earlierdrafts
of
We thankS. C. StearnsandP. Schmid-Hem- (2) r,ljmustdepend on an identifiableunique subset
thismanuscript.
us to see theirreplyto thispaper.
pel forallowing of phenotypicdimensionsforall j = 1,2, ..., m.
(3) rl,jmust be maximizedby a unique phenotype
((Plo, 7P2(o. *-'" (Pno))
(4) f, must be a strictlymonotonicallyincreasing
Appendix
functionof r%1 forall j = 1,2, ... m.
In orderto clarifythe conditionsunderwhichoptimiza- These statementsmaybe justifiedbythefollowingar-
tiontheorymightbe applicable to the structureand be- guments:
haviour of animals, we firstdefine three multidimen- Suppose thattwoseparateactivitiesj and k are notin-
sional abstractspaces, in whichindividualanimalsmay dependent,i.e. rlijand rlikboth depend in part on a
be representedas singlepoints: sharedsetofphenotypic variables,
say(cpi,,qPi, .., pqi).
(1) genotypespace, the dimensionsof whichrepre- Thus rij willdepend on ((pd, (Pei , m...,I (pqi) and
(P.i
sent different independentgenotypicvariables. rik willdependon ((p,i, (n,, .... q)qi, q,ri, ..., qv i)
(2) phenotypespace, then dimensionofwhichrepre- It is clear thatthe rewardsrliiand rlkcan be replaced
sent independentphenotypicvariables. The phenotype bya newreward, r,i,say, thatdependson ((Pdi, ei ...,
of the ith individualis representedas ((li, (P2i .... qpni)' (Pli, qPmi *..., (Pqi, (Pri. *... (Pvi).
Obviously the heritability of phenotypicvariables de- If r,jis consideredseparatelyfromrlikit may be pos-
pends on the nature of the mapping from genotype sible to predictthe phenotyethat maximizesr1i.This
space on to phenotypespace. phenotypewillcontainthe elements((po, (Pmo ..., (pqo,),
(3) performance space, thedimensionsofwhichrep- wherew denotesthevalue foreach phenotypicvariable
resentthe m different activitiesin whichthe animalen- whichmaximizesthe rewardrlj.It is extremelyunlikely
gages (m<=n). The location of an individualin each that the values for the shared phenotypicvariables
dimensionrepresentsthe rewardobtainedfromunder- whichmaximizethe rewardobtained will be identical
takinga particularactivity(e.g. the rate of food intake forboth activitiesj and k.
achieved while foraging). Thus, ifa constraint(the need to participatein activ-
An individualanimalmaybe characterizedby a point ityk) is invokedto explainwhyr1iis not maximizedas
in each of these spaces, and by a singlevalue fortotal predictedassumingthe independenceof j fromk, it is
reproductiveoutput. In thiscontextwe distinguishbe- inevitablethatrlikwill not be maximizedeither.Hence
tweenthreemeaningsof the word fitness: if k is a constrainton j, j will be a constrainton k.
(1) a synonymfor reproductiveoutput,withno ex- Suppose that the particulartype of individualthat
planatory content. producesthegreatestnumberof offspring can be identi-

116 OIKOS 49:1 (1987)

This content downloaded from 73.188.188.250 on Sun, 19 Oct 2014 19:51:38 PM

All use subject to JSTOR Terms and Conditions
fied,and thatsuch animalspossess the phenotype(qpl, References
p2Q , pnQ) Such animalsmustachieve maximumre- Charnov,E. L. 1976.Optimalforaging. The marginal value
wards fromall the activitiesin whichtheyengage. theorem. - Theor.Popul.Biol. 9: 129-136.
However,thereis no objectiveway of identifying the Cody,M. L. 1974.Optimization in ecology.- Science183:
rewardsbecause thereis no wayof dividingup whatthe 1156-1164.
Cowie,R. J. 1977.Optimalforaging in greattits(Parusma-
animal does into independentactivitiesthat involve jor). - Nature, Lond. 268: 137-139.
non-intersecting subsetsof the set of phenotypicdimen- Haldane, J. B. S. 1963. The acceptanceof a scientificidea. -
sions. There is, equally,no objectivewayof dividingup In: Weber, R. L. (ed), 1982. More randomwalks in sci-
the independentphenotypicdimensionsinto non-over- ence.Institute ofPhysics,Bristol.
Krebs,J.R. andDavies,N. B. (eds), 1984.Behavioural ecol-
lappingsubsets so thata particularsubset can be related
ogy,An evolutionary approach,Seconded. - Blackwell,
to a particularactivity,and therecan be no independent Oxford.
specificationof the rewardto be obtainedfromthe jth - and McCleery,R. H. 1984.Optimization in behavioural
the functional form of or ecology. - In: Krebs,J.R. andDavies,N. B. (eds). Behav-
activitynor for fli f2. iouralecology,An evolutionary approach,Second Ed.
It is however always possible to select sets of phe- Blackwell,Oxford,pp. 91-121.
notypiccharactersarbitrarily, determineempirically the -, Stephens,D. W. and Sutherland,W. J. 1983. Perspectives
values thatmaximizereproductiveoutput((pad, (PeQ,..., in optimalforaging.- In: Clark,G. A. (ed), Perspectivesin
and then circularly construct first-orderfitness ornithology, CambridgeUniv. Press,New York.
(pq,), Kuhn, T. S. 1970. The structureof scientificrevolutions.-
functionsand rewardsthatare maximizedby the same Univ. of Chicago Press, Chicago IL.
values forthe same set of phenotypiccharacters. Lewontin,R. C. 1978b. Adaptation.- Sci. Am. 239: 212-230.
Under the assumptionof optimalitytherewillexista - 1978. Fitness,survivaland optimality.- In: Horn, D. J.,
set of m objectivelydefinableactivities,performance in Mitchell,R. D. and Stairs,G. R. (eds), Analysisofecologi-
cal systems.Ohio State Univ. Press, Columbus, OH, pp.
each of whichwill be maximizedby natural selection: 2-31.
butthereis no possiblewayof identifying them.Even if Macnamara,J. M. and Houston, A. I. 1985. Optimalforaging
therewere therewould be no possibleway to relatethe and learning.- J. Theor. Biol. 117: 231-249.
rewardsto reproductiveoutput. Marris, G., Hubbard, S. and Hughes, J. 1986. Use of patchy
resourcesby Nemeritiscanescens (Hymenoptera:Ichneu-
The relationshipsbetweenphenotypeand first-order
monidae). I. Optimal solutions.- J. Anim. Ecol. 55: 631-
reward,and between first-order rewardand reproduc- 640.
tive outputcannot be empiricallyinvestigated(even if Maynard-Smith, J. 1978. Optimizationtheoryin evolution.-
they exist objectively) and thissuggests thatthe whole Ann. Rev. Ecol. Syst.9: 31-56.
Nagel, E. 1974. The structure of science. Problemsin thelogic
explanatoryscheme amounts to a reificationwith no of scientificexplanation. - Routledge and Kegan Paul,
contentthatis not circularlydefined. London, pp. 398-446.
The phenomenologicalrelationshipbetween pheno- Oaten, A. 1977. Optimal foragingin patches: a case forsto-
typiccharactersand reproductiveoutputmaybe repre- chasticity.- Theor. Popul. Biol. 12: 263-285.
Ollason, J. G. 1980. Learningto forage- optimally?- Theor.
sented Biol. 18: 44-56.
Pop.
- In Press a. Artificialdesign in naturalhistory:whyit's so
r2i = fs(q)li, (P2i , . (ni) easy to understandanimal behaviour.- In: Bateson, P. P.
G. and Klopfer,P. H. (eds), Perspectivesin Ethology.Vol.
7. Plenum Press, New York.
Rosenberg (1978) argues that the only possible map- - In Press b. Learningto foragein a regenerating patchyen-
ping, fs,fromphenotypeto reproductiveoutputis su- vironment:can itfailto be optimal?- Theor. Popul. Biol.
pervenience.Informallyexpressed,it is taken as a pre- Oster, G. F. and Wilson,E. 0. 1978. Caste and ecologyin the
mise thattwo animalsthatare physicallyindistinguish- social insects.- PrincetonUniv. Press, Princeton,NJ.
C. S. 1958. Adaptation,naturalselectionand be-
able possess the same fitness;but two different animals Pittendrigh,
haviour.- In: Roe, A. and Simpson,G. G. (eds), Behav-
may also possess the same fitness,shortcomings in dif- iour and evolution.Yale Univ. Press, pp. 390-416.
ferentaspectsoftheirphenotypebeingcompensatedfor Platt,J. R. 1964. Stronginference.- Science 146: 347-353.
-
in different ways. Because of this,he suggeststhatthe Pulliam, H. R. 1974. On the theoryof optimaldiets. Am.
Nat. 108: 59-75.
only way that phenotypicvariationcan explain vari- Pyke,G. H. 1984. Optimalforagingtheory:a criticalreview.-
ationsin reproductiveoutputis byenumerating thephe- Ann. Rev. Ecol. Syst. 15: 523-575.
notypicpropertiesof each of the organismsand asso- - , Pulliam, H. R. and Charnov, E. L. 1977. Optimal for-
ciatingwitheach set of propertiesthe reproductiveout- aging:a selectivereviewof theoryand tests.- Quart. Rev.
Biol. 52: 137-154.
put of the animal in question. Rosenberg, A. 1978. The supervenienceof biological con-
Such an exercisewouldlack explanatorypower,butit cepts. - Philosophyof Science 45: 368-386.
mightreveal regularitiesthatwould at least permitthe Ruse, M. 1977. Is biologydifferent fromphysics?- In: Colod-
identification of thephenotypeswiththegreatestrepro- ny, R. G. (ed.), Logic, laws and life, Some philosophical
ductivepotential.It would also avoid referenceto the complications.Univ. of PittsburghPress, Pittsburgh,pp.
89-128.
hypotheticalconceptualclockworkupon whichthe va- Sih, A. 1982. Optimalpatch-use:variationin selectivepressure
lidityof evolutionaryoptimizationtheorydepends. forefficient foraging.- Am. Nat. 120: 665-685.

OIKOS 49:1 (1987) 117

This content downloaded from 73.188.188.250 on Sun, 19 Oct 2014 19:51:38 PM

All use subject to JSTOR Terms and Conditions
Thompson, N. S. 1981.Towardsa falsifiable theoryof evo- tiqueof some currentevolutionary - Princeton
thought.
lution.- In: Bateson,P. P. G. andKlopfer,
P. H. (eds),Per- Univ.Press,Princeton,
NJ.
spectivesin ethology.Vol. 4, Advantagesof diversity. Williams,M. B. 1970. Deducingthe consequences of evo-
PlenumPress,NewYork.pp. 51-74. model.- J. Theor.Biol. 29: 343-
lution:a mathematical
G. C. 1966.Adaptation
Williams, selection.
andnatural A cri- 385.

Evolutionary shouldnotbe wasted

insights
ZoologischesInst. der Universitat,
StephenC. Stearnsand Paul Schmid-Hempel, 9, CH4051 Basel,
Rheinsprung
Switzerland

Summary. ofoptimiza-
In replytoPierceandOllason'scritique been discussedin the literature(e.g. Oster and Wilson
tiontechniques ingeneralandforaging
theory we
inparticular,
discussthelogicanduseofoptimality Mostoftheir 1978, MaynardSmith1978, McNeill Alexander 1982).
approaches.
argument is basedona misinterpretation
oftheunderlying logic
ofoptimization and,moregenerally,
theory ofbasictenetsof
method.
thescientific -
Weagreewithsomepoints notnewones 1. Definingfitness
- towardstheendoftheirlist- inparticularwithcertain
prob- Pierce and Ollason make heavygoingon severalpoints
lemsencountered whenanalyzingforaging behaviour.
where closer attentionto basic evolutionarytheory
would eliminatethe problemsbeforetheyarose. First,
there is the apparentlytroublesometerm, "fitness".
Pierce and Ollason state explicitly,in theirAppendix,
Introduction
thatiffitnessis takento meanreproductiveoutput,then
That thereare problemswithusingoptimizationtheory ithas no explanatorycontent.This is theirversionofthe
in ecologyand evolutionis no newsto theresearchcom- old accusation that evolutionarytheoryis circularbe-
munity.However, the precise sense in whichthe tech- cause it predictsthe survivalof the fittest,but defines
nique remains valid has not been appreciated in all "fittest"as those thatreproduceand survivemost suc-
quarters,as evidenced by Pierce and Ollason's (1987) cessfully.On the one hand, such remarksignore the
attack.They arrangetheirpointsin decreasingorderof broad palette of fitnessmeasuresthatare available for
importance,but forthe followingreasons we findour- help in makingpredictions.On the other,such remarks
selves in agreementwiththemonlyas theynear theend ignorethefundamentally circularnatureof deep axioms
of theirseries. in all branchesof science. We considerthe existingdi-
First,in analyzingsome problemsthatarisein model- versityof fitnessmeasuresfirst.
ling behaviour,theyhave generalizedtheircritiqueto In nature,organismsare born, reproduce,and die.
an attack not just on adaptationistthinkingin biology We observe the descendants,and in tryingto make
but on the generalrole of theoryin science. Even ifthe sense of the patternsthat we see, we inventabstract
situationin optimalforagingwere as bad as theythink- terms,like "naturalselection,""fitness,""adaptation,"
and it is not - theirstatementwould stillbe an unjusti- and so forth.These termshelp in explainingpatterns,
fiedexaggeration.Secondly,theirargumentis based on makingpredictions,and constructing consistentinter-
a misapprehensionof the claims of optimalitymodel- pretationsof observationsthatwould otherwiseappear
lers, whose conception of the organismand of con- to be unrelated.Thus thejustification foranydefinition
straintson the evolutionaryprocess is much more so- of the basic termsis practical.
phisticatedthan Pierce and Ollason are willingto ac- While a termlike "fitness"may be used loosely in
knowledge.There is not much gloryand oftenmuch evolutionarychat, in any specificmodel, whetherin
confusionto be gained in the destructionof a straw population genetics, optimal foraging,or life-history
man. theory,the termhas a concreteand quite unambiguous
We have chosen to replynot because theseissues are technicalmeaning.The meaningmayvaryfromfieldto
poorlyunderstoodby researchers,but because a pub- field,butin anygivencontextitis clear and, forthepur-
lished attackon optimization,especiallyone as flawed poses of theproblembeinganalyzed,itis rarelyifevera
as this,mightconfuseto the fieldby misrepresenting its circulardefinition.In populationgenetics,the meaning
accomplishmentsand its currentstatus. We begin our of fitness(usually W), is "that parameterbest repre-
commentsat a general level, then finishwith a dis- sentingdifferential reproductivesuccess in such a way
cussionof specificpointsraised by Pierce and Ollason. thatone can predictchangesin gene frequencies."Note
It is inevitablethat some of our points have already thatfitnessserves to help predictchanges in gene fre-

118 OIKOS 49:1 (1987)

This content downloaded from 73.188.188.250 on Sun, 19 Oct 2014 19:51:38 PM

All use subject to JSTOR Terms and Conditions