SAMPLE SIZES AND CONFIDENCE INTERVALS FOR WILDLIFE POPULATION RATIOS RAYMOND L. CZAPLEWSKI,1 WyomingGame and Fish Department, Cheyenne, WY 82002 DOUGLAS M. CROWE, WyomingGame and Fish Department, Cheyenne, WY 82002 LYMAN L. MCDONALD, Departments of Statisticsand Zoology, Universityof Wyoming,Laramie, WY 82071 One ofthemostcommontypesofdata used are not available untilthe data are at least in applied wildlifemanagementis the popu- partiallycollected.The computationsneceslationratio.These data are used to gauge re- saryfordetermination of whetheror not an productivesuccessand the effectof selective adequate sample size has been obtainedare harvesting of wildlife. rathertediousand do not lend themselvesto A commonproblemfacingwildlifeman- quick reviewwhilein thefield. how manyanimalsneed agersis determining A secondapproach,whichis the subjectof to be classifiedin orderto obtainan accept- thisdiscussion,considersthe individualaniable estimateof these populationratios. A mal as a unit.This procedureavoids mostof closelyrelatedquestionis the determination theabove-mentioned problems.However,the of confidenceintervalsaroundthe ratiosob- user mustbe willingto meet the following tained. requirements whichare listedin orderof imGiven a populationof animalsthereare 2 portance: generalsamplingstrategies forcollectingdata 1. The populationis clearlydefined. to estimatepopulationratios.One approach 2. The individualanimalsare sampled in a assumesthatthe area occupiedby the popurandomand unbiasedmanner.This is diflationcan be divided into quadrats (plotsficultto accomplishin applied manageperhapsbelt transects)and the quadratsare the problemcan onlybe addressed ment; randomly sampledwithina givenhabitattype, in thispaperas applicationwill superficially i.e.,thequadratis thesamplingunit.A survey vary with and habitat.Essentially species (census)of theanimalsis made on each of the therequirement is thateach animalin the sampled quadratsand the numbersof (e.g.) population has an and independent equal fawnsand does are recorded.In thiscase, the chanceofbeing"sampled."Extremeeffort populationratiois estimatedas usualwhileits shouldbe made in designof the studyto precisionis obtainedaccordingto "classical meet this condition.If the sampling is finiterandomsample theory,"see forexambiased, e.g., for"too many" does in estiple, Cochran(1977). Samplesize formulasexmation of a buck: doe ratio,thennaiveapist for the necessarynumberof quadratsto of plication themethodology presentedwill survey.However,in practice,sucha designis not matters. help The final ratios are still oftenimpracticalwhenfacedwiththereality biased! of time,budgetary,and planningconstraints. Also,the samplesize formulasdepend on es- 3. An upper limiton the populationsize is available,e.g.,themanagermustbe willing timatesof the sample varianceswhichoften to estimatethat thereare no more than N = 5,000 bucksand does in a population. 1 Present address: USDA Forest Service, Rocky Fortunately, sample size formulasare not Mountain Forest and Range ExperimentStation,240 West ProspectRoad, Fort Collins,CO 80526. sensitiveto errorsin thisvalue. Also,if the 121 122 Wildl. Soc. Bull. 11(2) 1983 (a ? b):100 estimatedupper limitis set too high,then the indicatedsample size is conservative (e.g.,[70 ? 8 fawns]/100does). (i.e., largerthannecessary)forthe desired precisionand confidence.The upper limit 1. Example for calculating adequate sample is forthepartofthepopulationunderconsizes and confidence intervals. sideration.For example,if a buck: doe raA wildlifemanagerneedsan Augustfawn: tio is desired,thenan upper limiton the antelopeherd.The totalnumberof bucksand does is required doe ratiofora pronghorn (total population minus the number of necessarynumberof does and fawnsto be classifiedmustbe determined.There are apfawns). 5,000 animalsin thisherdand it 4. In as faras possible,thesampleconsistsof proximately distinctanimals,i.e., the samplingis con- is feltthatat least3,600 are does and fawns. N equals 3,600in thefollowing. It Thisis more Therefore, ducted"withoutreplacement." thattherewill be about efficient than allowinganimalsto be "re- is also hypothesized sampled"and shouldbe a goal ofthestudy 80 fawns/100does. This will put thevalue of design.If the studyis conductedso thata "a" at 80. The desiredprecisionof the estiof theanimalscould matedratiois b = ?8 fawns/100does. Using proportion significant be classifiedmorethanonce,thentheoret- Table 1, underthecolumnheaded 80:100,we as one could find ?8.0:100 in the fifthrow which correicallythepopulationis infinite continuesampling"withreplacement"in- spondstocurvenumber5 in Fig. 1. UsingN = The figures and tablesin thispa- 3,600 in Fig. 1 and curvenumber5, the redefinitely. per can be used when animalsare resam- quiredsamplesize is about825. That is,a few to morethan800 does and fawnsshouldbe claspled by takingtheresultscorresponding the largestpopulationsizes reported(i.e., sifiedto insure(with90% confidence)thatthe N = 12,000 is essentiallyinfinitewith re- fawn:doe ratio will have precisionof ?8 fawns/100does. spectto theformulas). In a similarmanner,Table 1 and Fig. 1 can of the procedureto be deJustification be used in reverseto obtainan approximate scribedis containedin the appendix where confidenceintervalon the estimatedratio.In formulasare derivedthatwill providemore the above example,supposethe sample size withinthetaexactanswersthaninterpolation actuallyobtainedis n = 1,200 and the comblesand curvespresented.As a generalguideputed ratiois 70 fawns/100does. The point line,all resultsare developedat the90% con- n = 1,200 and N = 3,600 is close to curve fidencelevel(see theappendixforotherlevels number6. EnteringTable 1 withtheratio70: of confidence). 100 and curve(row) 6, the precisionis ?5.7: 100. In conclusion, thefawn:doe ratiois withESTIMATING SAMPLE SIZE AND in (70 ? 5.7 fawns)/100does withapproxiCONFIDENCE INTERVALS mately90% confidence. Populationratiosare oftenexpressedin the formof young/100femalesor males/100fe2. Example for adequate sample sizes with males. This expressionof a ratiowill be demore than 2 classes. notedas "a: 100." A ratioof70 fawns/100 does wouldtherefore be expressedas 70:100,where Continuingexample 1, supposethe buck: a = 70. Usingthisnotationforexpressing pop- doe ratiois also of interest. Withthehypothulationratios,the end pointsof the 90% con- esized ratioof 80 fawns/100does we expect fidenceintervalare denotedby: 1,600 fawns,2,000 does, and 1,400 bucksin WILDLIFE I 3,000 000000000 000000000 co co _ co O CD t 0 CO + +1+1+1+1 +1 +1 --l+1 _1 +1 _ OC _ . _CO O OC ill +- +- 7l 11-l RATIOS * Czaplewski et al. POPULATION 1 l 2,500 1 COO +-l +l 2,000 t ,T 00 0 0 t0 00 co 0 C 00 0c0 00 0t C6 50,500 6 +l +l +l +l +l +l --+ 1,000 X a) -a >) - 4 . O .. _ .. .. .. .. x LooO --+ .. .. .. I'l co co 500 +l +l + +l +l+ 0 ._ +1 ~~~+1 Ca N a) -+1 +1+lll 000000000 -z - C CI cc o- .. . .. . .. .. .. Ct! Lo .. .. .. co cq CtI . g aO0 0) c? ~~~~+I+I +l +l+ + +l +l ++l +I +l +I +l+1+I ~+l ~~~+1 ut k o CQ 000000000c Co C 0)~~~~~~~ 4.^ En t _ +1 +1 +1 +1 +1+1 +1 + C, L 0000000 _ C0 CZ +l +l +l +l +l +l +l +l ~~~+l ct o C) c Lt 0 0 '_ 6 X es m=5 Q) X o~~~~~ 4000 6000 POPULATION SIZE 8000 K3300 12,000 C o .~~~~o _________i~6dcC, Q. O3 o o C)t 00 2000 Fig. 1. Samplesize (n) as a function of population size (N) of 2 age or sex classesand a givenlevelof precision indicatedin Table 1 (90%confidence). C) = 5 +1 + ~~+1 C) 123 CS: 00 CIO CO C 't C, _ _ ~~~+1 +1 +1 +1 +1 +1 +1 +1 +1 ~~+l+l o Lo +l o Go om o. +l+l+l ooo +lI+l+l the populationof 5,000 animals.Also,about 367 fawnsand 458 does shouldbe in thesample of 825. The hypothesized buck: doe ratio is 1,400:2,000or 70 bucks/100does and suppose the desiredprecisionis ?8 bucks/100 does. Usingthecolumnheaded 70:100 in Table 1, ?8.0:100 is foundtocorrespond tocurve number4 on Fig. 1. Also N = 1,400+ 2,000 = 3,400 in thiscase. From Fig. 1, about 690 bucks and does should be sampled,and we expectto obtainabout284 bucksand 406 does given the hypothesizedratio 70 bucks/100 does. Taking the largerof the 2 values for the does (e.g., 458 does in example 1) the adequate size forthecombinedsampleis: n = 367 fawns+ 284 bucks+ 458 does = 1,109animals. About 1,100 totalanimalsfromthe populationof 5,000 shouldbe sampled. MINIMUM AND MAXIMUM SAMPLE SIZES a) = Fig. 2 containsa simplified versionofTable 1 and Fig. 1. It is intendedthatFig. 2 be used Wildi. Soc. Bull. 11(2) 1983 124 PRESPONDING TO CURVE 20 be selected.The samplesize would be larger in bothcases. POPULATION RATIO 100 A ?4: 100 8 ?3:100 C ?2: 100 40 50 :1:00 ?67100 70 7 91 ?4:100 ?5:100 43:100 0100 100 i00 ?7: 100 5:100 ?00 13:100 t1I: 1X 4. Example for calculating maximumrecommended sample size. ? 7:100 2POO~~~~~~~~~~~SZ We desireto estimatethe postseasonbull: cow ratio of an elk herd. We estimateapproximately 5,000 bulls and cows are in the herd. UsingcurveC fromFig. 1, a realisticmaximumsample size is approximately 1,625. If 1,625bullsand cowsare classified and thebull: So. cow ratioapproximates 20:100,thenthe90% confidenceintervalfromthe top of Fig. 2 wouldbe (20 ? 2):100. "Tighter"confidence Fig. 2. Recommended samplesize (n) as a function intervalsrequire larger sample sizes. Howof populationsize (N) of 2 age or sex classesand a givenlevelofprecision indicatedin theinserted table ever,increasingthe sample size above 1,625 willnotyieldmeaningful in the (90% confidence). improvements precision.To see this,the sample size 1,625 to curve7 in Table 1. Increasing corresponds by fieldpersonnelas a generalguidelinefor the sample size to 2,750 (curve9) yields(20 minimumsamplesize,maximumsamplesize, ? 1.1):100. The workloadmustbe approxiand the correspondingprecisionassociated matelydoubledin orderto improvethe prewitha givencombinationof samplesize and cisionto ?1.1 bulls/100cows. theestimatedratioat approximately 90% confidence. .DOC~~~~~~~~~~~~ 1,5001 1,000 ' v _ 500an EZ- 0 2000 4000 6000 8000 10,000 12,000 POPULATION SIZE SAMPLING "WITH REPLACEMENT" Ifeveryanimalhasan equal chanceofbeing classified(reclassified)on each samplingoccasion,thentheoretically thepopulationis inIt is desiredto estimatethepreseasonfawn: finite(N = oo). Figures1 and 2 extendonly doe ratioin a mule deer herd. Currentestito N = 12,000. However,the curvesare almatessuggestapproximately 3,000 does and mostflatin thisregionand N = 12,000 can fawnsare presentin the population. safely be used in place of N = x0 in all but Usingcurve A fromFig. 2, the minimum themostprecisestudiescorresponding tocurve numberof does and fawnswhichwe recom9 in Fig. 1. In this extremecase (curve 9), mendforthesampleis about530. If thefinal samplesizes in the neighborhood of 4,000 to fawn:doe ratio is approximately70 fawns/ are necessaryto obtaintheprecisionin5,000 100 does,theapproximate90% confidenceindicatedin thebottomrowof Table 1. tervalis ?9:100, i.e., (70 ? 9 fawns)/100does (see the tablereproducedon Fig. 2). 5. Example for calculating recommended Assumingthe samplingprocedureis ransample size when sampling with replacedom and unbiased,thereis a 90% chancethat ment. the truefawn:doe ratioof the herd lies between61 fawns/100does and 79 fawns/100 In a "quick and dirty"census,severaldifdoes. If one desiresa smallerconfidencein- ferentindividualsare to cruisea networkof terval,thencurveB or C fromFig. 2 should roads over severalweeksand classify(or re3. Example for calculating minimumrecommended sample size. WILDLIFE POPULATION RATIOS * Czaplewski et al. 125 classify) everyantelopetheysee. Assuming that thiscan be done in a randomand unbiased manner,how manydoes and fawnsshouldbe classified(reclassified)in orderto achieve a precisionof ?8.0 fawns/100does? Also assume thatundersimilarconditionsthispopulationhas had a fawn:doe ratioof about80: 100. Using Table 1, curve 5 is selectedand fromFig. 1 a sample size of n = 1,000 doe to N = 12,000. and fawnantelopecorresponds Because curve5 is almostflatin thisregion,a sample of about n = 1,000 is adequate fora doeseventhough precisionof ?8.0 fawns/100 From Fig. 2, animals are being reclassified. our minimumrecommendedsample size is about n = 600 and the maximum recommendedsamplesize is about2,000. As in the otherexamples,the processcan be reversedto obtainapproximate90% confidenceintervals,i.e., use N = 12,000 in the figures. assumea maximumof N = 2,000 males and femalesand a desiredprecisionof ?2 males/ 100 females.Becausethenecessary samplesize decreasesas the hypothesized ratiodecreases (see Example6), thecolumnheaded 10:100in Table 1 is recommendedand will give a conservative answer.Usingcurve4 and N = 2,000 in Fig. 1, a samplesize of 600 is largerthan necessaryfor precisionof ?2.0:100 if the "true"ratiois 5:100 (90% confidence). We recommendthatthe left-handcolumn headed (10:100) in Table 1 be used for all hypothesized ratiosbelow 10:100. Afterdata are collectedand a computed ratiobelow 10:100 is obtained,say 8:100, the precisionindicatedin the firstcolumnof Table 1 is conservative. For example,withn = 1,000animalssampledout of N = 2,000total, n = 1,000is abouthalf-waybetweencurves6 and 7 in Fig. 1. Usingthefirstcolumnof Table 1 (10:100)theprecision is betterthan? 1.3: 100. The confidence interval (8 ? 1.3 males)/ SPECIAL CASES 100 femalescan be reportedwith>90% con6. Exampleto be usedforincreasesin thehy- fidence. pothesizedratio. Considerthe fawn:doe samplingproblem 8. Exampleto be usedforhypothesized ratios in Example 1, except that the hypothesized above 100:100. ratiois 100 fawns/100does.UsingthelastcolIf a ratioof 125 fawns/100does has been umn of Table 1 we mustdrop to row 6 to obtained (for the given herd under similar achieve a precisionof ?8.0 fawns/100does. conditions in previousyears),we are againout Hence in Fig. 1 we mustmove up to curve6 of therangeof Table 1. We recommendthat and therequiredsamplesize corresponding to the manager use the last column (100:100) N = 3,600 is approximately n = 1,125. when planningforsample sizes in thiscase. The sample size mustincreasefromabout Theoretically, theroleofdoesand fawnscould 800 to about 1,100 in order to achieve the be reversedto obtain80 does/100fawnsand same precision(?8:100). This is a generalrethecolumnheaded 80:100 used to determine sult.As the hypothesized ratioincreased(left sizes. However,most managerswill sample to rightin Table 1) the usermustdropdown express the ratioas fawns:doe and the larger in the table in orderto achievea fixedprecisizes sample suggestedby thelastcolumnare sionand henceup in Fig. 1 to a largersample recommended. As an alternative, themoreexsize. act formulasin theappendixcan be used. We 125 fawns/100doesand suppose hypothesized 7. Exampleto be usedforhypothesized ratios N = 2,000 is an upperbound on the number below10:100. of does and fawnspresent.For precisionof Considera sample problemwherethe hy- ?8:100 and usingthecolumnheaded 100:100, pothesizedratioofmales/females is 5:100.Also curve 6 in Fig. 1 is suggested.This yieldsa 126 Wildl.Soc. Bull. 11(2) 1983 recommendedsamplesize of about900. Suppose the data are collectedwithn = 900 and a ratioof 110 fawns/100does is computed.As a trickto obtaina confidenceinterval,express 110 fawns/100does as 91 does/100fawns.Usbetween80:100 ing row 6 and interpolating and 100:100,theprecisionis about ?7.3 does/ 100 fawns (90% confidence)-or (91 ? 7.3 does)/100fawnswhichyields83.7 does/100 fawnsto98.3 does/100fawns.Theselastratios convertback to 102 fawns/100does to 119 fawns/100does (90% confidence).The precisionis almost(110 ? 8 fawns)/100does. CONCLUDINGREMARKS Table 1 and Fig. 1 are presentedto allow thesamplesize whenestimating interpolation necessaryfora statedprecisionand hypothesizedratio(90% confidence). Also,theycan be used to allow interpolationwhen obtaining approximate90% confidenceintervalsfor a computedpopulationratio. Figure2 containsour recommendation for minimumand maximumsamplesizesand the levelofprecision.It is intended corresponding thatFig. 2 be used by fieldpersonnelduring data collection. Hypothesized ratiosmustbe availableto use the tablesand figures.Such valuesmay come froma preliminarysample, prior data collected in the same area undersimilarcondiconsiderations tions,theoretical (e.g.,theratio of male fawnsto femalefawnsmightbe assumedto be 100:100 forplanningpurposes), themanager'sintuition, etc.Clearlywithmore precise input information, adequate sample size is estimatedwithmoreprecision. The estimatedsamplesizesare notsensitive to errorsin theestimatedpopulationsize. The curvesin Figs. 1 and 2 are relativelyflatfor populationsexceeding500 animals. The requirementfor a random and unbiasedsamplingprocedureis difficult to meet. This is especiallytruewhen membersof the classesoccupydifferent segmentsof the hab- itator whenmembersof 1 classare morevisible thanmembersof the otherdue to plumage, feedinghabits,tendencyto runin herds, etc. It will testthe manager'sknowledgeto overcomesuchproblemsand insureunbiased data collectiontechniques.A secondproblem, whichis not quite so serious,is thathidden correlations may be introducedintothe data whenclusters(herds,all animalson a quadrat, will not etc.) are classified.These correlations causebiasedratiosbutthestatedprecisionmay be incorrect(too high or too low). No good guidelinesexistin the literatureforthe solutionof thisproblem(in as faras the authors are aware). If the precisionis betterthanthe value claimed by the theorybehindthispaper, thenthe manageris conservative and is usinga samplesize largerthannecessary.In the othercase, his sample size is not large enough. We recommendthe methodology presentedin this paper as a very general guidelineforadequate samplesize whenever the manageris confidentthat his sampling procedureis unbiased.For studieswherethe exactprecisionof theratiosmustbe known,a professional statistician, knowledgeablein finitesamplingtheory,shouldbe consulted.A variationof quadratsamplingor clustersamplingwill likelybe recommended. SUMMARY Adequate sample sizes are given for the numberofanimalstobe classified in a random and unbiasedstudyforestimationof populationage and sex ratios.The desiredlevel of precisionand confidence can be specified.Advantagesof thisprocedureare: (1) the data and precisionare summarizedin standard forms(e.g., 40 ? 3 males/100females),(2) the tablesand curvesare easy to use in the field,(3) minimumand maximumsamplesizes can be recommended, and (4) the tablesand curvescan be used in reverseto obtainapproximateconfidenceintervalson the estimatedpopulationratio. WILDLIFE POPULATION RATIOS * Czaplewski et al. Acknowledgement.-Thispaper was originally developed as PlanningReport6-A by the PlanningSectionof the WyomingGame and Fish Department,Cheyenne,WY 82002. Now Z 1- (1 (1 A2 COCHRAN, W. G. 1977. SamplingTechniques,3rd ed. JohnWiley& Sons,New York.413pp. Received26 July1982. Accepted25 September 1982. lOO(P - c) = (l 1P) and LITERATURE CITED looP -A 2 + C(l lOO(P + c) (1 lO0c p)2 - (6) - P) lOOP (P+C) - (P -C) lO~c loom P)2 1- 127 C(l (l-P) - P)' (7 For an approximatesymmetricprobabilityinterval on a:100, i.e., (A + B):100, take = APPENDIX lO loom (8) of N animalswhereP population Considera finite valuebetween(A - A,) and (A2- A) inthefirst ofthepopulation an intermediate category is theproportion in thesecondcate- inequations(6) and (7). Solvingequation(8) forc and and Q = 1 - P is theproportion ofanimalssampledwithout substituting into(3) together with(4), we obtain gory.Let n be thenumber and za, be the"2-tailed"valuefromthe replacement zZa2AN For nP 2 5.0 and nQ - 5.0 normaldistribution. .N (9) value to10:100,theminimum corresponding (roughly B2100(N -1) probability in Table 1), a 100(1 a)% inconsidered tervalon p is givenby (100 + A)2 Givenequation(9), therelationships in Table 1 and Figs.1 and2 weredevelopedbya computer program. Also,givena hypothetical valueforA:100,thedecategory siredprecision?B, thepopulation inthefirst proportion wherep is theobserved size N and thede(Cochran1977:51).To have precisionof ?c forthis siredlevelofconfidence, 1 - a, (i.e.,za) theadequate we can solve interval samplesizecan be computedby(9). Finally,afterthedata are observedand A is estimatedbya, equation(8) can be written in theform (2) c, PQ\n(N-1) p +Z" P(N-n) n(N -l) (1) b forn Za(a + lOG0) 10 a(N -n) n(N - 1) 0 Nz2PQ to obtainan approximatesymmetric 100(1- a)% confidence interval on a:100. If the samplingis conducted"withreplacement" to the formof thenequation can be can be transformed The proportions usedwithN = oc.Dividing (9) ratios(a:100,[A ? B]:100,etc.)bythefor- thenumerator population anddenominator by N, itis easytosee mulas thatequation(9) willconverge to looP A zaA(100 + A)2 100 + A' (1-P)' (11) 100B 100 loop (4) Similarly,equation (10) convergesto 100+ A' (Ic2(N - 1) + z2pQ (3) Giventhe probability p ? c, the resulting interval, ona:100wouldbe from interval A1:100toA2:100where 100(P - c) A + 100) V10 a/n. 9. Example for the use of equation (9). 1 - (P - c) 1-0(P + c) b -zj(a (5) Supposethehypothesized buck:doe ratiofora postseasonmuledeerherdclassification is 30 bucks/100 128 Wildl.Soc. Bull. 11(2) 1983 does.We estimate aboutN = 2,000bucksand doesin the herd.For 95% confidence, zo5= 1.96 and from (9) theadequatesamplesizeforprecision of?B ?3 bucks/100 doesis n= ( (1.96)2(30)(2,000) [?3]2100[1,999] + (1.96)230 [100+ 30]2 230,496 [106.46 + 115.25] = 1,040bucksand does. 10. Example for the use of equation (10). Thereare 109 bucksand 402 does observedin an unbiased,randomizedpreseasonantelopeclassifica- tion.The herdratiois 109:402or27.1bucks/100 does. Therefore, a = 27.1, n = 109+ 402 = 511,and we estimatethataboutN = 2,000bucksanddoesareinthe preseason herd.If90%confidence isdesired, zl = 1.645 and fromequation(10) 1.645(27.1 + 100) 10 27.1(2,000 - 511) 511(1,999) = 20.9(0.1988) = 4.2. Therefore, thereis (approximately) a 90%chancethat thetrueherdratiofallsin theinterval (27.1 ? 4.2 bucks)/100 does. COYOTE PREDATION AVERSION WITH LITHIUM CHLORIDE: MANAGEMENT IMPLICATIONS AND COMMENTS RICHARD J. BURNS, U.S. Fish and Wildlife Service, Predator Ecology and Behavior Project, Utah State University,UMC 52, Logan, UT 84322 Conditionedtaste aversionhas been proposed as a methodto detercoyotes(Canis latrans) frompreyingon sheep. The method involvesplacing muttonbaits laced with a strongemetic,lithiumchloride(LiCl), on the range. Coyotessupposedlyfindand eat the baits,becomeill,and subsequently avoidsheep because theyassociatethesheepwithsickness et al. 1974,1976,Ellinset al. 1977). (Gustavson Attempts also have been made to controlcoyote predationon turkeys(Meleagrisgallopavo) usingLiCl-laced turkeycarcasses(Ellins and Catalano 1980). Someinvestigations priorto 1978castdoubt on theeffectiveness of usingLiCl baitsto deter coyotesfromkilling(predationaversion) and,in a summary paper,Griffiths et al. (1978) concludedthatno judgmentcouldyetbe made regardingthevalueofLiCl in preventing coyote predation.In thispaper I synthesizethe literature on thesubjectsince1978,discussthe mostprobablemanagementimplication(assumingLiCl producedpredationaversionin coyotes), and pointout somepresentinformationneeds. In defending an earlierpublication, Conoveret al. (1979) maintained thatmoreresearchwasneededonpredation aversion with LiCl, whereasGustavson (1979)feltthatthe existing studiesdemonstrated the successof themethod.Cornelland Cornely(1979)believedthatLiCl fedtocoyotes in a variety of foodsdiscouraged potentially dangerous coyotesfromsoliciting foodat a campground. Burns(1980b)demonstrated thatLiCl's salt flavor interfered withtheability ofcoyotes to formaversions tobaitsandpreykilling. Burns and Connolly(1980) thenproduceda measurableblack-tailed jack rabbit(Lepus californicus) baitaversion withlowdosesofLiCl thatappearednotto provideflavorcues to coyotes. The baitaversion, however, wasnot transferred topredation aversion; coyotes continuedto killjackrabbits. Ellinsand Martin (1981)laterfoundthatcoyotescoulddetect LiCl and avoidbaitsat concentrations below