Measures of the Amount of Ecologic Association Between Species
Transcripción
Measures of the Amount of Ecologic Association Between Species
Measures of the Amount of Ecologic Association Between Species Author(s): Lee R. Dice Reviewed work(s): Source: Ecology, Vol. 26, No. 3 (Jul., 1945), pp. 297-302 Published by: Ecological Society of America Stable URL: http://www.jstor.org/stable/1932409 . Accessed: 26/05/2012 07:11 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 [email protected]. Ecological Society of America is collaborating with JSTOR to digitize, preserve and extend access to Ecology. http://www.jstor.org MEASURES OF THE AMOUNT OF ECOLOGIC BETWEEN SPECIES LEE ASSOCIATION R. DICE Universityof Michigan In many ecologic studies thereis need to expressin a quantitative mannerthe species are degreeto whichtwo different associated in nature. Nevertheless,no such measurehas up to the presenttime come into generaluse by ecologists. Of themeasuresofassociationbetween species previouslyproposed,the "coefficient of association" of Forbes ('07) is useful,thoughno estimateof its statistical reliabilityhas in the past been emofassociation" ployed. The "coefficient of Michael ('20) is too complicated for generaluse. The "association value" of Hacker ('21) varies with the total numberofassociationsbetweenall the species in a given locality and thereforeis impracticable formost fieldstudies. Van Deventer ('36) has calculated the percentage of association between certain species of birds,but his briefstatement gives no details of the methodof computation. Recently Edmondson ('44) applied the chi-squaretest as a measure of the deviationfromchance expectationof the occurrencestogether of two given species of rotifersin a seriesof lakes, but he did not use any measureoftheamount of association betweenthe species. The coefficientof association developed by Forbes ('07) considersthe numberofassociated occurrencesoftwo given speciesin a seriesofrandomsamplescompared to the numberof such joint occurrencesexpected by chance. In orderto we must therecalculate this coefficient fore count the number of samples in whicheach of the two species occursand also thenumberofsamples in whichboth species occur together. Let us indicate by a the numberofsamples in whichspecies A occurs either alone or together with species B, by b the numberof samples in which species B occurs,by h the numberof samples in whichboth species occur,and by n the total numberof samples examined. The chance of species A occurringin any particularsample in a/n. Similarly,the the seriesis therefore chance ofspecies B occurringin any sample is b/n. The chance that species A and B will both occur in any given sample is a/n X b/n. The numberof samples in the series in which species A and B would be expected to occur togetheris thereforea/n X b/nX n or abin. To we now divide the obtain the coefficient number of samples (h) in which both species are observed to occur by the numberof samples (abin) in which they would be expected to occur by chance: hn h -. Coefficient of association = a-= abln ab' This formulais exactlythat proposedby Forbes, except that the symbols are different. A coefficientof association of 1.0 shows that the two species under considerationoccur togetherin exactly the number of sample units expected by chance. A coefficientsmaller than 1.0 shows that they occur togetherin fewer samples than would be expected by of0.5 chance. For example,a coefficient shows that they are foundtogetheronly halfas frequentlyas expectedby chance. Similarly, a coefficientlarger than 1.0 indicates that the species being compared are associated more frequently than would be expectedby chance alone. The coefficientof association therefore gives a readilyunderstandableindication of the degreeto whichthe association between any two species conformsto expectationon the basis of chance. of associaThe value of the coefficient tionbetweenany two givenspecies varies with the abundance of the formsin the area studied. If the two species com- 297 298 LEE R. DICE Ecology,Vol. 26,No. 3 pared are both abundant, both will ap- with species A (B/A) would be 30/40 pear in a highproportionof the samples = 0.75. This indexshowsthaton threetaken. The chance of both species oc- quartersof the plots on whichspecies A curringtogetherin any given sample is occurred,it was associated with B. If thereforehigh,and it will be impossible we furtherassume that in the series of forany greatdeviationfromexpectation sample plots under consideration the to occur. Under these conditions the number (b) of plots in which species B coefficient of associationcan neverbe far appeared was 30, then the association from 1.0. On the other hand, if both index of species A with species B (A/B) species are rarein the situation,thenthe would be 30/30= 1.0. This indexshows chance of theiroccurringtogetherin any that on all the plots wherespecies B apgivensample is also low. Any deviation peared species A also was present. fromexpectation may result in a very It willbe noted thattheassociationinhighor verylow value forthe coefficient dex may differdependingon whichspeof association. A highcoefficient there- cies is used as the basis of comparison. foredoes not necessarilymean that the Such a difference may point out an imtwo formsare stronglyassociated, nor portantecologicrelationbetweenthetwo does a low coefficientalways indicate species. Frequently one species is dethat the formsare stronglyrepulsed. A pendent upon another, without there coefficientof 1.0 shows that the two being any reciprocaldependency. It is formswereobservedto occur togetherin believedthat the associationindicesmay exactly the numberof samples expected sometimes be of value in discovering by chance, but it does not necessarily such unilateraldependencies. indicatethat the formsdo not tend to be Because the reciprocalassociation inassociated. The coefficientof associa- dices are differentit will be important tion is thereforenot a measure of the always to indicate which species is used amount of association between species, as thebasis ofcomparison. This may be but onlya measureof the amount of the done by placing the name of the base deviation of the numberof theiroccur- species second in the statement. Thus, rences together from the number ex- for a theoreticalexample: "association = 0.01." pected by chance. index gracilis/bairdii To providea simpleand directmeasure In some ecologic studies it will be deofthe amountofassociationbetweenany sirableto use a measureof the amountof given species (A), taken as the basis of associationbetweenspeciesthat does not comparison,and any otherspecies (B) I change depending on which species is hereproposethe associationindex. This used as a base. Such a measure, here indexis obtainedby dividingthe number called the coincidenceindex,has a value (a) ofrandomsamplesof a givenseriesin intermediatebetweenthereciprocalassowhich species A occurs into the number ciation indicesA/B and B/A. Thus: (h) of samples in whichspecies A and B occur together. Therefore: Coincidenceindex= 2h a+b Associationindex B/A = h/a. In theexamplepreviouslygiventhe coinThe formulaforthe reciprocalindex is: cidenceindexwouldbe (2 X 30)/(30 + 40) = 60/70 = 0.86. The coincidenceindex Associationindex A/B = h/b. thereforeis a measure of the amount of For example,ifspecies A should occur associationbetweenboth the two species on a total of 40 sample plots (a = 40), compared. The values of the association indices and if species B should occur together with A on 30 of these plots (t = 30), and of the coincidenceindex range from then the association index of species B 1.0, which indicates association of the July,1945 AMOUNT OF ECOLOGIC ASSOCIATION two species in all the samples examined, to 0.0, which indicates complete failure of associationunderthe conditionsof observation. Intermediatevalues give the proportionalamountofassociation. The easy to comprehend. indicesare therefore No indicationof a possible deviation fromchance expectation is given either by the association index or by the coincidenceindex. It will thereforeoftenbe desirable to give one or other of these of associaindicesand also the coefficient tion. All ofthesemay be calculated with only slighteffortfromthe same data. No statisticaltest of the reliabilityof any of these measuresof association has been devised. However, the chi-square test (Snedecor, '40: 16-19) will show whetherthe combinationsof species in the samples of a particular series may possiblybe due to chance errorsin random sampling. This test should thereforealways be applied to the data from whichany association index,coincidence index, or coefficientof association is derived. In applying the chi-square test it is necessary to determinethe number of unitson whicheach possiblecombination of the two species would be expected to occurby chance. The expectednumber ofsamples in whichtwo species,A and B, would be expectedto occur togetherhas already been shown to be abin. Species A wouldbe expectedto occuralone in the number of samples from which it has been recorded(a) less thenumberofsamples in which it is expected to occur togetherwith B. Similarly,B is expected to occur alone in b samples minus the numberofsamplesin whichB is expected to occur along with A. The chance of speciesA not occurringin any givensample in the seriesis (n - a)/n. Similarly the chance of B not occurringin any sample is (n - b)/n. The chance that neitherA nor B will occur in a given sampleis therefore(n - a) /nX (n - b)/n and thetotal numberofsamplesin which neitherspecies is expected to be present will be (n - a)/n X (n - b)/nX n or (n - a) (n - b)/n. From the deviations BETWEEN 299 SPECIES fromthese expected numberschi-square is calculated. As an example let us assume a seriesof 100 samples in whichspecies A has been foundalone in 5 samples,species B alone in 20 samples, both species A and B togetherin 20 samples,and neitherspecies in 55 samples (table I). Then the value Theoreticalexample of the association betweentwospecies (A and B) shown by 100 samples TABLE I. Number of samples Class Observed A only A+B B only Neither Totals Expected 5 20 20 55 15 10 30 45 100 100 Deviation from expecta- (deviation)2 expected tion -10 +10 -10 +10 0 6.667 10.000 3.333 2.222 22.222 a = 25; b =40; n = 100; coefficient of association 2.0; association index A/B =0.5; association index B/A = 0.8; coincidence index= 0.62; degree of freedom=1; X2=22.222. = of a is 25, b is 40, and n is 100. The numberof samples in whichA and B are expected to occur together by chance of associa(abin) is 10. The coefficient tion is accordingly20/10 or 2.0, which showsthat the two speciesin theexample occur together twice as frequentlyas would be expectedby chance. The association index of species A with species B (A/B) is 20/40 = 0.5, and of species B with species A (B/A) is 20/25 = 0.8. The coincidence index is (2 X 20)/(25 + 40) = 40/65 = 0.62. In calculating the chi-square deviations in our example we used data from two species. The numberof degrees of freedomis thereforeone. The total chisquare is 22.222 (table I). From a table of the values of chi-square (Fisher, '36, table III; or Snedecor, '40, table 9.1) it is evidentthat this amount of chi-square with one degree of freedomis far above the one per cent. level of significance. The associationofspecies A and B on the sample plots must thereforebe due to some factorother than chance errorsin random sampling. We may conclude 300 LEE R. DICE that thereis a tendencyin this situation forspeciesA and B to be ratherstrongly associated in their distribution. From the differencein association indices, moreover,it is suggestedthat species A is somewhat more stronglyassociated with B than species B is with species A. The chi-squaretestis not reliablewhen the numberofexpectedunitsin any class fallsbelow five. If possiblethe expected number in each class should be ten or more. Should the number of samples available fora particularstudynevertheless be fewand if therebe only a single degree of freedom,Yates' "correction" (Snedecor,'40: 168-169) may be applied. It is better,however,in everyfieldstudy oftheamountofassociationbetweenspecies to securea sufficient numberofsamples so that no class willcontainless than ten expectedunits. The magnitudeof chi-squarevaries to some degree with the abundance of the species compared. Two stronglyassociated species mightoccur togetheron a large number of sample plots and yet, if both were abundant, the deviation fromexpectationmightbe slightand chisquare thereforesmall and not statistically significant. Also, two species that have only a slighttendencytoward association, if both were rare in the samples, might occur together frequently enough to have a large and very significant chi-square. Care should therefore be taken not to interpretthe magnitude of chi-square as indicatingthe strength of association betweenthe species under consideration. On the other hand, the degree of significanceof chi-squaremay be an indirect but valid measure of association. The amount of association between threeor morespecies can be determined by means of the coincidenceindex and coefficient of associationby an expansion of the formulas used for two species. Also the significanceof the deviation fromexpectationofthe observedamount of association betweenthese species can be determinedby the chi-square test. Attemptsto measure the association of Ecology,Vol. 26,No. 3 threeor morespecies will,however,very rarelybe practical. The numberofsamples needed to securean adequate measure of the associationof only two species is very considerable. A fully reliable measureof theamountofassociationbetweentwo species will seldom be secured with much less than 100 samples, and morethan thisnumberwillnearlyalways be desirable. It willrarelybe possibleto secure the much greaternumberof comparable samples needed to measure in one step the amount of association between threeor morespecies. So farwe have discussedonlythe association between two or more given species. We may also need to measurethe amount of association betweena species and someinanimatefeatureofitshabitat. For this purposewe may use the indices of association or coincidence,the coefficient of association, and the chi-square test of significance. We may thus compare the occurrenceof a given species withthe simultaneouspresenceof rocks, logs, trees,clumpsofbushes,or any other featureoftheenvironment. It mustnot be expected, however,that the calculation of thesestatisticswillalone solve all theproblemsoftherelationshipsbetween organismsand theirenvironments. The abundance of the individuals of either of the species concerned on the sample unitsis not consideredin the calculation of the coefficient of association or of the indices of association or coincidence. Such an omissionfrommeasures of associationhas been criticizedby Calvert ('22). Furthermore,no account has been taken,in thesecomputations,of the possible tendencyof the individuals to be associated with one another in pairs, families,or largersocial aggregations. The abundance and also the social behavior of the individuals of any given species obviously affect the frequency withwhichthe formappears in any kind of sample. Therefore the frequency with which any species appears in a given set of samples includes, to some degree, an indication both of its abun- July,1945 AMOUNT OF ECOLOGIC ASSOCIATION' dance and of its social behavior. So long as any given sample is as likelyas any othersample to containa representative of each of the species concernedwe may use these sample counts as a basis for computingthe association between the species. We will readilygrant,however, that these measuresof association apply onlyto the samplestakenand that if the two species involved should occur in otherdensitiesor ifone or both should have another type of social grouping, such as mightbe truein otherseasons of values forthese the year, then different measures of association would probably be obtained. No-methodshave yetbeen devised forpredictingthe values of these measuresof associationunderotherpopulation densities or under other social groupingsthan those exhibitedby a particularsample series. The samples that are used as a basis forthe measurementof association may be any of the usual sample units employedin ecologicfieldwork. They may be quadrats, transects,or sample plots of any kind, or they may be sample catches obtained by sweeps of an insect net, secured by drags of a tow net or dredge,or collected in any other standardized manner. The essential feature is that the two species to be compared actually must have been in close associationwithone anotheriftheyare recorded in the same sample unit. Any sample unit that covers a large area or that includes several different types of ecologic communities,therefore,will not usually be a proper base for the calculations. Sample unitsbased on timeintervalswill be admissibleonly if theyalso are space unitseach of whichcoversan area of not too great extent. The size of the units fromwhich the samples are obtained influencesall these measures of association. This is true because the frequencywith which any species appears in a series of samples is dependentnot only upon its densityand its social tendencies but also upon the size of the area fromwhichthe sample is drawn. If the sample plots are of large BETWEEN SPECIES 301 size, almost all the species in the local biota, except only the rarer ones, will appear on a largeproportionof the plots. On sample plots ofsmall size, on the contrary,even the most abundant species are likelyto appear on only a small fractionoftheplots. For thisreasonheterogeneous samples, such as the collections of fishcomparedby Forbes ('07), do not providea sound basis forthe calculation of the amount of association between species. Inasmuch as the value of all these measures of association varies with the size of the sample units employed,it is evident that the sample units of the serieson whichany ofthesecomputations is based must all be of the same size. The size of sample unit to be employed must of course be adjusted to the abundance and also to the size and mobilityof the species being compared. The most efficientmeasure of association seems generallyto be obtained when the more abundant one of the species under consideration occurs on about half of the sample unitsin the series. All sample unitson whichestimatesof association are to be based should be taken at random,using this termin the statisticalsense. If any selectionis used in locatingthe sample units the method of selection should be fully described. Samples taken systematically,such as one collectionin each ofa seriesof stands or of communities,usually cannot be used properlyas a basis for statistical calculations. Any measureofassociationis of course true only for the particular place and time for which it is calculated. The ecologic relations between organisms vary fromtimeto time,fromcommunity to community,and fromplace to place. It will always be important,therefore, to specify the particular conditions upon which any given measure is based. It will usually be impossible to combine data taken at different times,in different places, in different typesofcommunities, in differentmanners,or fromdifferent sizes of samples to serve as a basis-for 302 LEE Ecology,Vol. 26,No. 3 R. DICE the calculation of the degree of association betweenspecies. The fact that two species may tend to occur together,as shownby a highassociationindex,coincidenceindex,or coefficient of association, or by a significant chi-square deviation from expectation, does not in any way explain the reason fortheirassociation. It may be thatthe two formsare mutuallyattracted to one anotheror that one formis in some way dependent upon the other. The two species may, however,be associated only because of the presence in their habitat of certainfeaturesattractiveor essential to both. In such a case the two associated species may have no relationship at all to one another,otherthan selection of the same habitat. These statisticsat best are a measure only of the actual amount of association between two species ratherthan of the amount of attraction betweenthem. SUMMARY The coefficient of associationof Forbes indicates the amount of association betweentwo givenspecies comparedto the amount of association betweenthem expected by chance. In orderto providea simple direct measure of the amount of association of one species with another the associationindex is proposed. If a is the number of random samples of a givenseriesin whichspeciesA occursand h is the number of samples in which anotherspecies B occurstogetherwithA, then the association index B/A = h/a. Similarly,ifb is the numberof samplesin whichspecies B occurs,then the association index A/B = h/b. There is also proposed a coincidenceindex,2h/(a + b), whose value is intermediatebetweenthe two reciprocal association indices. As a measure of the statisticalreliabilityof the deviation shown by the samples of a given series fromthe amount of association expected by chance, the chi-square test may be used. LITERATURE CITED Calvert, Philip P. 1922. Methods for expressing the associations of different species. Ecology 3: 163-165. Edmondson, W. T. 1944. Ecological studies of sessile Rotatoria. Part I. Factors affecting distribution. Ecol. Monog. 14: 31-66, 5 figs. Fisher, R. A. 1936. Statistical methods for research workers; Sixth edition. Edinburgh and London: Oliver and Boyd. xiii + 339 pp. Forbes, S. A. 1907. On the local distribution of certain Illinois fishes:an essay in statistical ecology. Bull. Illinois State Lab. Nat. Hist. 7: 273-303, 15 maps, 9 pls. Hacker, H. P. 1921. On the distribution of anophelines in the BrickfieldsRoad area of Kuala Lumpur, November, 1917. Federated Malay States Malaria Bureau Reports 2: 3-28, 1 map. Michael, Ellis L. 1920. Marine ecology and the coefficientof association. Jour. Ecol. 8: 54-59. Snedecor, George E. 1940. Statistical methods applied to experimentsin agricultureand biology. Ames, Iowa: Iowa State College Press. xiii + 422 pp., Illus. Van Deventer, William C. 1936. A winter bird community in western New York. Ecology 17: 491-499, 3 figs.