Evolution, 34(2), 1980, pp. 332-341

INTERSPECIFIC COMPETITION, ISLAND BIOGEOGRAPHY AND NULL HYPOTHESES

P. R. GRANT AND I. ABBOTT Department of Ecology and Evolutionary Biology, University of Michigan, Ann Arbor, Michigan 48109, and Department of Soil Science and Plant Nutrition, University of Western Australia, Nedlands 6009

Received April 6, 1979. Revised October 8, 1979

We conducted a field study of some of Darwin's Finches. Finally, we draw at­ Darwin's Finches (Geospiza species) in or­ tention to some unsolved problems in bio­ der to assess the relative importance of in­ geography, concerning principally the terspecific competition and habitat fea­ separation of potentially conflicting effects tures in determining the observed of different processes such as dispersal and biogeographic, ecological and morpholog­ competition. ical characteristics of these species (Abbott et aI., 1977). Strong et al. (1979) have crit­ Why Their Methods are icized one of our methods and have rean­ Unsatisfactory alyzed a small portion of our data. They Strong et al. (1979) generated expected employed stochastic models to generate ratios of beak sizes among birds on the expected beak size differences between California Channel islands, the Tres Ma­ sympatric species, and then compared ex­ rias islands and the Galapagos. For the pected with observed differences. Finding first two groups of islands they used a a generally close correspondence between computer to draw randomly the observed expected and observed differences, they number of species on each island from concluded that random processes are suf­ within families of birds on the adjacent ficient to account for the observations, and mainland. They repeated the exercise 100 that therefore there is no need to invoke times to obtain an estimate of sampling deterministic processes such as competi­ error. For the Galapagos they chose tion as we had done. Strong et al. (1979) species at random, then a population from obtained the same results and drew the one of the islands at random for each of same conclusion from analyses of beak the species already chosen, until the num­ size differences among birds on the Tres ber of species populations matched the Marias islands of Mexico and the Channel observed ones on all of the major islands. islands of California. Simberloff and his Ratios were calculated separately from associates have also drawn the same con­ observed species combinations and ran­ clusion from a series of other analyses per­ domly paired species, and then compared. formed in like manner (Connor and Sim­ There are several problems with these berloff, 1978; Simberloff, 1978). procedures. In several ways the analyses We take issue with the procedures tend to maximize the risk of making a Strong et al. (1979) have used in their Type II error, that is, favoring acceptance analyses and with the way in which our of the null hypothesis when it is false. statements and interpretations have been 1. The first problem is that the choice represented. We identify five problems in of species within families is unfortunate. their analyses and five sources of confu­ Species of birds in different genera are less sion in the interpretation of results. We likely to be in the same feeding group or find no evidence in their analyses or ar­ guild than are species in the same genus, guments to change our previous conclu­ and hence are less likely in general to be sion that interspecific competition has in potential competition for food (excep­ played a role in the adaptive radiation of tions are easy to conceive, and observe, 332 INTERSPECIFIC COMPETITION 333

Grant, 1966). We know of no systematic ence in results stems from a difference in study undertaken to test this assertion, but the species pool. We consider our species it is a view commonly held on the basis of pool to be appropriate and theirs to be observations and from the reasoning that inappropriate. the more closely related two species are This point is of general importance. the more they will have similar adaptive Strong et al. (1979) could find no reason systems for locomotion, food gathering, to reject the null hypothesis for the total breeding, etc. (e.g., Darwin, 1859; Hair­ Tres Marias and Channel island data. Yet ston, 1964; Lack, 1965). the beak sizes of members of the same A simple illustration can be given with feeding guild suggest that competition has Darwin's Finches. Species in the genus influenced part of the structure of the Tres Geospiza show strong similarities in feed­ Marias avian community (Grant, 1966). ing methods, locations and actual diet The only biological justification for us­ (Snodgrass, 1902; Lack, 1945, 1947; Bow­ ing the family level in treating the data man, 1961, 1963). Our study provided a that we can find is in Simberloff (1978, p. quantitative confirmation of this similari­ 714)-"Still more important than the low ty. Species in the genus Camarhynchus visibility of interspecific competition in also show strong similarities in feeding obscuring its relationship to biogeographic methods, etc., but differ markedly from distributions is the fact that putative com­ all Geospiza species on the one hand and petitors are rarely pairs of taxonomically from Certhidea olivacea on the other (ref­ [congeneric?] and morphologically similar erences above). Pooling all species from species." This claim is unsubstantiated, these three genera, then choosing pairs at which should therefore serve as a stimulus random to examine beak ratios, is a mean­ for further research. An analytical justi­ ingless exercise as a test of competitive fication is that the family level of analysis effects because there is no a priori reason is less sensitive to specific taxa than is the to believe that unrelated species would genus level (Strong et al., 1979). Be that compete, except under the most extreme as it may, resorting to the family level of view that diffuse competition noticeably analysis runs the risk of throwing the baby affects all species in a community. We out with the bath water; or, to be more suggest that it may be worse than mean­ correct, drowning the baby by using too ingless; it may obscure real competitive large and deep a tub. effects within genera by diluting the phe­ 2. The second problem is in the choice nomenon to the point of disappearance. of a mainland area for comparison with Caswell (1976), Colwell (1979) and Inger the island (Grant, 1966). Geographical and Colwell (1977) have made a similar variation in community membership and point. beak sizes of the members within the Grant (1980) has shown with a multi­ mainland region could influence the re­ variate analysis of beak dimensions that sults of a random simulation of the process populations of sympatric pairs of Geospiza of island colonization. species are not a random sample of all In an earlier analysis of the species/ge­ possible paired populations of those nus ratio of island bird communities, Sim­ species; observed pairs tend to differ more berloff (1970) showed that observed ratios than randomly chosen pairs. Character tended to be higher than those expected displacement is a possible, but clearly not from the random sampling of mainland necessary or sufficient, explanation (Grant, species, and certainly not lower as would 1979). In some of their analyses Strong et be expected from an hypothesis of com­ al. (1980) used Geospiza and Camarhyn­ petition (Grant, 1966). With regard to the chus species together, and found that ob­ problem of correctly identifying the source served sympatric assemblages of species pool of species on the mainland, he wrote, (pairs, triplets, etc.) can be considered "A study currently in progress indicates random samples. We suggest the differ- there is little geographical variation in S/G 334 P. R. GRANT AND I. ABBOTT

[species/genus] values for land birds, par­ islands is strongly influenced by the equi­ ticularly when localities of similar latitude probability assumption. We suspect that are compared" (Simberloff, 1970, p. 27). other expected biogeographic, ecological The source pool for the California Chan­ and morphological properties of island nel islands and Tres Marias islands prob­ populations are similarly affected. This is ably comes from areas more extensive lon­ not intended to belittle the attempts to gitudinally than latitudinally. One come to grips with a very complex prob­ therefore wonders how errors of choosing lem; only to point out that the problem the correct source pool might influence the has not yet been solved, and the interim results of species/genus and beak differ­ report is not convincing. ence analyses using the stochastic model 4. The fourth problem concerns the approach. Given this uncertainty, and for generation of expected beak size ratios for maximum biological realism, the safest archipelago islands (e. g., Galapagos) procedure is to use the species in an ad­ when there is no identifiable mainland jacent mainland area equal in size to the source area and species pool. The solution islands (Grant, 1966; Abbott, 1975), rath­ adopted by Strong et al. (1979), Connor er than all those in a whole country such and Simberloff (1978) and Simberloff as Mexico (Simberloff, 1970). The chances (1978) in similar analyses, is to treat the of detecting nonrandom colonization are archipelago as a universe from which in­ greatly reduced by using excessively large dividual islands are populated. Beak ra­ mainland species pools when the main­ tios are calculated (Strong et al., 1979) land communities vary geographically in from randomly combined populations; a systematic way. then the results of repeated random pair­ 3. In the original analyses Simberloff ings are compared with the actual data. (1970) assumed equiprobability of dispers­ Our criticism here is that the two sam­ al among all species in the mainland pool, ples tested for similarity (the real data is in order to obtain expected species/genus one sample, and the set of randomly gen­ ratios on islands. This assumption seems erated faunas is the other) are not inde­ unlikely to be true (Grant, 1970; Terborgh pendent. In fact, the real data for any giv­ et aI., 1978), as has been recently ac­ en island are a subset of the randomized knowledged, "Biologically this assump­ data. The statistical bias that this kind of tion implies that all species have equal nonindependence causes is not random; it dispersal and persistence abilities. In fact, consistently favors the acceptance of the species have different abilities to disperse null hypothesis, and hence maximizes the and persist, and this assumption is there­ risk of making a type II error, since the fore absurd" (Connor and Simberloff, two samples are artificially made more 1978). similar by the inclusion of the same data Our purpose here is to point out that in each. until a biologically realistic stochastic Presumably the justification for their model can be constructed, with allowance procedure, following explicit statements in made for different dispersal abilities, the Connor and Simberloff (1978) and Sim­ observed species/genus ratios and "com­ berloff (1978), is that the method of com­ munity-wide" beak ratios of birds on is­ bining populations at random "does not lands cannot confidently be ascribed solely uniquely determine" the observed ratios, or predominantly to random processes. "and so the tests are not circular" (Connor When equal weight is given to species and Simberloff, 1978). If not circular they with low or zero probability of dispersal are at least elliptical, for all populations and to species with high probability of of the Geospiza species on the 15 islands dispersal, the analysis is tipped in favor they considered are sympatric with at least of accepting a null, random, hypothesis. one other species, that is, all species pop­ Simberloff (1978) has shown that the ex­ ulations enter the calculation of expected pected number of species shared by two ratios and all provide the observed ratios INTERSPECIFIC COMPETITION 335

pare the data. First, as mentioned above, w U the random sampling assumes equal dis­ Z 200 w .. a: persal abilities for all populations of all w u, "- ISO species; this is unlikely to be even approx­ Ci imately correct. Second, observed and ex­ Z 100 t""······ "Observed pected ratios are not independent. Third, < "'-f . w there is a tacit assumption that the birds ~ 50 -- ~l' have fixed properties, that is, no in situ '- Predicted evolution, for the pattern of beak ratios 4 can be simulated simply by randomly NUMBER OF SPECIES combining existing populations without FIG. 1. The average difference between size requiring further change in beak dimen­ neighbours in beak size and shape in relation to the sions. A "no-evolution" assumption leaves number of species on an island. Differences are Ma­ hanging the question of why populations halanobis D2 values, calculated on the basis of beak of the same species differ in beak size depth, length and width for each pair of species (Grant, 1980). To generate the predicted curve, among islands in the first place, and why species were randomly combined in pairs, triplets, no two sympatric species differ by less quartets and quintets, and arithmetic mean D2 val­ than 15% in at least one beak dimension ues (horizontal lines) and 95% confidence limits (ver­ (Grant, 1975, 1980). Are we to attribute tical bars) were calculated from samples of SO. Data this complex pattern of variation to in Abbott et aI. (1977) and Abbott et al. (unpubl.) were used in the calculation of the observed differ­ chance? Doubtless a stochastic model ences. could be constructed which gives results approximating observed variation (d. Raup and Gould, 1974), but would the with which the expected ratios are assumptions be any more acceptable? In compared. The calculation of dispersal sum, the random model is patently un­ abilities from observed frequency-of-oc­ realistic. currence distributions (Connor and Sim­ These difficulties are compounded by berloff, 1978; Simberloff, 1978) is subject the test procedure used by Strong et al. to the same criticism. (1979). They tested the distribution of ob­ 5. Their analysis of just the Geospiza served minimum beak size ratios above data alone is the most similar to ours. and below the mean of the null commu­ They obtained a negative correlation be­ nity by a 2-tailed binomial test. This is not tween expected average minimum ratios the most powerful test to use because it and number of species, and concluded takes no account of the known magnitude that "a negative correlation is expected of deviations of observed from expected from randomly assembled communities, values; the same criticism applies to some so finding a negative correlation among of their other analyses. Moreover a real communities does not show character I-tailed test is appropriate because unidi­ displacement" (Strong et al., 1979, p. rectional deviations from random expec­ 900). We agree that although a competi­ tation are predicted from a competition tion hypothesis predicts a negative corre­ hypothesis. These are important points lation, it is also to be expected from a ran­ because from their Figure 1 it appears that dom (noncompetition) hypothesis. This is the observed mean for every sample is because a rank-ordered sample is like a greater than the null mean. This trend, if broken stick-the more breaks (large sam­ correct, is consistent with a hypothesis of ple) the smaller the pieces (ratios of adja­ competition. cent bill sizes). The problem then is to dis­ We have performed a multivariate ver­ tinguish between the two hypotheses. sion of the Strong et al. analyses, and this The task is not easy because there are has produced grounds for rejecting a ran­ three objections to the construction of a dom hypothesis and accepting a competi­ simple random model with which to com- tion hypothesis (Fig. 1). Observed beak 336 P. R. GRANT AND 1. ABBOTT

ratios tend to be larger than those expect­ displacement, an evolutionary process). ed from randomly combined species. The Strong et al. (1979, p. 897) point out the three objections listed above apply to our distinction in their introduction, but then analysis as much as to the analysis of go on to use the term character displace­ Strong et al. Note, however, that statis­ ment to include both processes. This is tical evidence partly favors a competition confusing because it is inconsistent with hypothesis despite a bias towards accept­ other specific definitions of character dis­ ing the null (random) hypothesis. Without placement as an evolutionary process complete confidence in the method of (e.g., Grant 1972a). analysis, we conclude that beak ratio data Thus Strong et al. (1979, p. 899-900) give evidence of competition among Geo­ wrote, "Abbott et al, (1977, p. 164, figure spiza species. 12) have concluded character displace­ Strong et al. (1979) performed one fur­ ment to be common in Geospiza." Else­ ther test with the Geospiza data. We sug­ where we are represented as claiming gested that competition may have stronger "that there is greater character displace­ effects on islands with fewer species than ment on islands with fewer Geospiza on islands with many (Abbott et aI., 1977). species" (p. 900), "infer[ring] a version of Strong et al. (1979, p. 901) reasoned from community-wide character displacement this that there should be "a tendency for among the Galapagos finches" (p. 908), islands with fewer species to fall farther and that we "assumed character displace­ above expected" [in the analysis discussed ment" (p. 909). above] than those with more species. This In fact, we claimed to have found evi­ was tested by comparing observed minus dence of interspecific competition. Our expected beak ratios with number of data did not allow us to choose between species per island. As expected from the explanations involving the very different competition hypothesis the sign of the cor­ processes of differential colonization and relation coefficient was negative in two character displacement (sensu Grant, analyses with Geospiza data (beak length 1972a). The choice can only be made and beak depth) but the correlations were where 'mainland' source populations and not statistically significant. They conclud­ colonization routes are readily identifiable ed, "If character displacement does oper­ (Grant, 1969). Within the Galapagos ar­ ate through minimum ratios of sympatric chipelago it is rarely possible to trace out finches, this sort of community-wide anal­ colonization routes with confidence (Grant, ysis cannot detect it" (p. 901). We agree 1967). Bearing this in mind we drew a with the conclusion, but for a different cautious conclusion about character dis­ reason. With only two degrees of freedom placement (sensu Grant, 1972a). "So by in each correlation analysis, a near-im­ this indirect argument [see Abbott et aI., possible demand is being made of the 1977, p. 176, for details] we infer that data. some morphological and ecological char­ acter displacement probably occurred, Other Misrepresentations and Confusions either in the initial stages of differentiation 1. There is confusion over the term or during the subsequent establishment of character displacement. As has been sympatry with the closest relative, al­ pointed out before (Grant, 1969), an en­ though we cannot say in which particular hanced character difference, such as a cases." large difference in bill size between two In short, we claim that interspecific sympatric species on an island, can be competition has occurred among Darwin's brought about by competition in two Finches, and that character displacement ways; by the elimination of intermediate­ was the consequence in some unidentified sized species (differential colonization, an instance. ecological process) and by an evolutionary 2. "The assumption that insular coexis­ response of one or both species (character tence begets greater sympatry than conti- INTERSPECIFIC COMPETITION 337

nental coexistence is of untested general competitive exclusion by congeners, un­ validity, but it is common in the ecological suitable structural habitat, absence or literature (Abbott et al., 1977). We would scarcity of preferred or suitable foods, or suppose that geometrically congruent inability to cross to some islands suffi­ sympatry among species sometimes but ciently often or in large enough numbers not always occurs on islands, and that it (chance colonizations)" (p. 169; see also is less frequent on mainlands" (Strong et Grant, 1966). al., 1979, p. 909). The second sentence in the quotation We neither referred to the assumption from Strong et al. (1979) implies that our in other work nor made it in our own as analyses were inadequate. We have al­ we were not concerned with making com­ ready given reasons for considering ours parisons between islands and mainland. adequate and theirs inadequate. Here we Regardless of how one defines "geometri­ take the opportunity to point out that our cally congruent sympatry," there is em­ conclusions rest on many more analyses pirical evidence that most of Darwin's than the one contested by Strong et al. Ground Finches (Geospiza species) on the (1979); these involved numbers and bio­ same island live in the same environment mass of finches, diets in relation to beak and encounter each other (Gifford, 1919; morphology, morphology in relation to Lack, 1945, 1947; Bowman, 1961; Abbott abundance, distributions and combina­ et al., 1977): they coexist in sympatry. tions of species on islands, habitat diver­ 3. "As an example of our reasoning, we sity and predation, all of which were used disagree in principle with Abbott et al. to test traditional null hypotheses. (1977), who conclude that food supply and 4. In an earlier analysis, bill size dif­ interspecific competition have jointly de­ ferences between congeneric species of termined the patterns of evolution and birds were found to be more frequently species diversity of the Galapagos Finch­ larger on the islands of West Indies than es. If their analyses of competition and on the nearby mainland, to a statistically habitat were adequate, only a portion of significant extent (Grant, 1968). Strong et the evolution and ecology of the finches al. (1979) were unable to duplicate these has been determined by these factors" results, and found the trend to be present (Strong et al., 1979, p. 910). but not statistically significant. We in turn Since we did not say that the two factors have been unable to duplicate the results solely determined the patterns, these re­ of Strong et al. (1979). In view of the un­ marks are as misleading as they are un­ certainties about the most appropriate necessary. Some investigators set up a mainland source pool with which to com­ problem in such a way that all variation pare the islands, we tried different areas in a phenomenon can be attributed to two of mainland (e.g., North and Central or three factors and to no others (e.g., Amercia, Central America alone). The re­ Terborgh, 1971). We did not adopt this sults were the same as in the original anal­ circumscribed approach. In the Introduc­ ysis; we see no reason to alter the original tion we stated, "We report results of field conclusion. Since the original data are studies undertaken to investigate some published in Schoener (1965), it should be factors responsible for inter-island differ- possible for someone neutral to the argu­ ences in ground-finch morphology and ment to settle it. ecology" (p. 152; italics added). The rest 5. "Although character convergence is of the Introduction makes it clear why predicted from some models of community these particular factors were chosen for integration (Schoener, 1969; Cody, 1973), detailed analysis. We also made the nec­ we agree with Hespenheide (1973) that essary statement that many factors affect convergence and displacement are quite complex phenomena, for example, "Gaps opposite phenomena. The fact that both [in the distribution of a species] such as are predicted from competition theory these could result from chance extinctions, might render the theory a catholicon, es- 338 P. R. GRANT AND I. ABBOTT pecially if the theory cannot predict which with logical primacy over other hypothe­ should occur in specific instances" (Strong ses, that other hypotheses must first be et al., 1979, p. 908). tested against, but that is rarely consid­ Two points may be made in reply to ered at all by ecologists. This is the null this. The first is that the equilibrium the­ hypothesis that community characteristics ory of island biogeography itself might are apparently random" (Strong et al., similarly be viewed as a "catholicon" (cf. 1979, p. 910). Abbott and Grant, 1976), as it encompass­ The logical primacy of randomness is es both equilibrial states and nonequilib­ debatable. Where different causal factors rial states which will presently or even­ are implicated in the determination of tually lead to an equilibrium (MacArthur complex phenomena like community and Wilson, 1967). Despite this disadvan­ structure, it is just as valid to test con­ tage, it has led to much productive re­ trasting deterministic explanations against search on mechanisms and processes be­ each other as it is to test each one against cause its elements can be tested, and it has a random hypothesis. However, regard­ not yet been replaced by a more obviously less of logical primacy, the use of non-in­ falsifiable theory. teractive models in community studies The second point is that there is a logic holds much promise (e.g., see Caswell, which connects divergence and conver­ 1976; Hubbell, 1979), in two respects. gence. Both are predicted from competi­ Such models should help to clarify the im­ tion theory. They are alternative solutions portance of randomness itself, while at the to basically the same competitive problem same time they are likely to stimulate in different ecological circumstances. more rigor in attempts to demonstrate the Moreover the most comprehensive set of operation of causal factors, as has hap­ models specifies the different conditions pened in population genetics through the under which each result is expected (Wil­ neutralist-selectionist debate. To elabo­ son, 1975). Furthermore, Wilson (1975) rate the first point, randomness in the null has pointed out that one particular ob­ hypothesis used by Strong et al. (1979) is served convergence, involving copepods only partial, for the components of the studied by Rigler and Langford (1967), hypothesis-species and their attributes­ conforms to the model. This is not to say have been supplied by nature; it is their that all convergences and divergences are relations which are random. So the im­ brought about by competition (Grant, portant question is, how much have ran­ 1972a). A judgement has to be made on dom processes shaped communities? the possible role of competition, on the If the history of studies of population basis of evidence from geographical vari­ cycles are a guide to what we might an­ ation, feeding ecology, etc. Nor does this ticipate in community studies, the intro­ say that divergence and convergence are duction of random models (d. Cole, 1951, equally likely to be brought about by com­ 1954) will take the mystique out of mac­ petition; divergence seems generally the roscopic phenomena, such as log-normal more likely (see Grant, 1975; Wilson, distributions of species abundances, and 1975). force critical attention upon the mecha­ nisms and deterministic processes that DISCUSSION produce repeated, biologically interpret­ One of the main themes in the papers able, patterns and events (Christian and by Connor and Simberloff (1978), Simber­ Davis, 1964; Chitty, 1967; Krebs and loff (1976, 1978) and Strong et al. (1979) Myers, 1974; Lidicker, 1975). Anotherles­ is that there is a need for testing hypoth­ son which may be learned from popula­ eses involving competition or any other tion studies is that random patterns are no deterministic process against appropriate guarantee of random causes, for Hassell null hypotheses. Their attitude is sum­ et al. (1976) have shown that deterministic marized, "We propose another possibility models of single populations can yield ran- INTERSPECIFIC COMPETITION 339

dom fluctuations of numbers in time (see Instead, the results of competition may also Gilpin, 1975). be manifested by which particular species Our criticisms have not been directed in a genus are present on an island, and at the need for appropriate null hypothe­ their bill sizes and associated characteris­ ses, with which we firmly agree, but at tics. These manifestations were shown to the use of a particular randomization pro­ be consistent with an hypothesis of com­ cedure to construct these null hypotheses. petition in a study of the birds of the Tres The artificiality of the random 'scram­ Marias islands (Grant, 1965, 1966). Six­ bling' model and in some instances its de­ teen congeneric pairs of species could oc­ pendence upon the data that are used to cur on these islands, as deduced from the test it, reduce its acceptability. pattern of occurrence of species among Another problem in biogeography and genera on the mainland and the occur­ evolutionary biology which is illustrated rence of genera on the islands. But only by this debate is the need for critical tests two pairs actually occur on the islands. that will allow confident rejection, as op­ These are the second and fourth most dis­ posed to nonacceptance, of a particular similar pairs in bill length. The occurrence hypothesis. For example, consider the hy­ of two such dissimilar pairs by chance pothesis that the low species/genus ratio seems unlikely, although attaching prob­ of birds on islands is determined in part abilities to so few occurrences is difficult. by interspecific competition (Grant, 1966). However, an independent test of the com­ Simberloff (1970) and Abbott (1975) have petition hypothesis was made with more shown that observed ratios are actually extensive data from the West Indies, and higher than expected from an hypothesis it gave a similar result; the competition of random sampling of mainland species. hypothesis was supported by the way in On the surface this would appear to be which a null hypothesis was rejected not just nonacceptance of the competition (Grant, 1969). hypothesis but a clear rejection of it, be­ Where there is little or no scope for dif­ cause departures of observed from ex­ ferential dispersal among genera we pected are in the opposite direction from would expect competitive effects to pre­ those predicted by the competition hy­ dominate. This may be the case on the pothesis. Aside from the difficulties with Galapagos. Thirteen of 15 islands have the random model discussed earlier, there lower species/genus ratios of birds than is a hidden complication. Species within expected by a random sampling procedure a genus may have similar propensities to (Connor and Simberloff, 1978). The au­ disperse and colonize islands. If these are thors did not test the null hypothesis that greater within some genera than others, as many islands have higher ratios than as seems likely to us and to Simberloff expected as have lower ratios than ex­ (1970), the species/genus ratio on islands pected. It is rejected by a I-tailed Sign test could be in equilibrium between two op­ (P = .011). There are other interpreta­ posing processes-differential dispersal tions of this result (Strong et al., 1979; see among genera, tending to elevate the ra­ also Connor and Simberloff, 1978), but tio, and intrageneric competitive exclu­ the point is that the competition hypoth­ sion, tending to lower it. If so, the ob­ esis cannot be ruled out. served higher-than-expected ratio provides Such conflicting processes complicate no evidence for a process, competition, analyses of biogeographic patterns (Sim­ which has actually operated, because berloff, 1978). While the refinement of sto­ another process, differential dispersal, has chastic models may help to resolve these been of over-riding importance. Therefore complications, we propose that detailed the competition hypothesis has not been ecological studies hold out better hope, adequately tested and rejected by com­ particularly if experimental and if directed paring species/genus ratios on islands and towards specific questions of competition mainland. and dispersal (Simberloff, 1969, 1976; 340 P. R. GRANT AND 1. ABBOTT

Simberloff and Wilson, 1969; Grant, The first two objections also apply to 1972b). predictions of island bird properties from a knowledge of mainland birds. Some ACKNOWLEDGMENTS confusions over the nature of character We thank A. Dunham and B. R. Grant displacement, the degree of sympatry on for helpful comments on the manuscript, islands and the relative importance of in­ and D. Simberloff for allowing us to see terspecific competition are dealt with. manuscripts before publication. To R. K. Current problems in biogeography in need Colwell we are specially indebted for clar­ of attention are (1) separating the poten­ ifying the statistical issues and providing tially conflicting effects of different pro­ an enlightening analogy between size ra­ cesses, such as competition and dispersal, tios and a broken stick. and (2) devising tests which permit clear rejection, as opposed to nonacceptance, of SUMMARY an hypothesis. We conclude that stochas­ tic models may be a useful tool in hypoth­ Some of the evidence for interspecific esis testing in biogeography, but until they competition among Darwin's Finches (Ab­ can be made realistic their usefulness will bott et al., 1977) has been challenged by be severely limited. Strong et al. (1979). These authors were able to predict the degree of morphological differences actually observed among the LITERATURE CITED finch species from stochastic models. They ABBOTT, 1. 1975. Coexistence of congeneric species concluded it is unnecessary to invoke de­ in the avifaunas of Australian Islands. Austral. terministic processes such as competition. J, Zool. 23:487-494. We argue that their methods of analysis ABBOTT, 1., L. K. ABBOTT, AND P. R. GRANT. 1977. Comparative ecology of Galapagos Ground are seriously flawed by a lack of realism Finches (Geospiza Gould): evaluation of the im­ and by biases that tend to favor accep­ portance of floristic diversity and interspecific tance of the null hypothesis, hence their competition. Ecol. Monogr. 47:151-184. conclusions are unacceptable: ABBOTT, I., AND P. R. GRANT. 1976. Non-equilib­ rial bird faunas on islands. Amer. Natur. a) All species within a family were 110:507-528. compared, even though some are so dif­ BOWMAN, R. I. 1961. Morphological differentiation and adaptation in the Galapagos finches. Univ. ferent ecologically that competition be­ Calif. Berkeley Publ. Zool. 58:1-302. tween them is, a priori, extremely un­ --. 1963. Evolutionary patterns in Darwin's likely. An analysis by genera as was finches. Occas. Pap. Calif. Acad. Sci. 44:107­ done for Geospiza, or by feeding guild, 140. is more meaningful. CASWELL, H. 1976. Community structure: a neu­ tral model analysis. Ecol. Monogr. 46:327-354. b) All populations of all species were CHITTY, D. H. 1967. The natural selection of self­ assumed to have equal dispersal abili­ regulatory behaviour in animal populations. ties and equal chances of reaching all Proc. Ecol. Soc. Austral. 2:51-78. islands in the archipelago; this is ex­ CHRISTIAN, J, J" AND D. E. DAVIS. 1964. Endo­ crines, behavior and populations. Science tremely unlikely to be even approxi­ 146:1550-1560. mately true. CODY, M. L. 1973. Coexistence, coevolution and c) A tacit assumption of no-evolution in convergent evolution in sea bird communities. the construction of random models Ecology 54:31-44. leaves hanging the question of why pop­ COLE, L. C. 1951. Population cycles and random oscillations. J. Wildlife Manage. 15:233-252. ulations of the same species differ in --. 1954. Some features of random population beak morphology among islands, and cycles. J. Wildlife Manage. 18:2-24. why no two congeneric, sympatric, COLWELL, R. K. 1979. Toward a unified approach species differ by less than 15% in at to the study of species diversity, p. 75-91. In J, F. Grassle, G. P. Patil, W. Smith, and C. least one beak dimension. Taillie (eds.), Ecological Diversity in Theory and d) Expected and observed ratios of Practice. International Co-operative Publishing beak size were not independent. House, Fairland, Maryland. INTERSPECIFIC COMPETITION 341

CONNOR, E. F., AND D. SIMBERLOFF. 1978. nae: a study in variation. Occas. Pap. Calif. Species number and compositional similarity of Acad. Sci. 21:1-159. the Galapagos flora and avifauna. Eco!. Monogr. ---. 1947. Darwin's Finches. Cambridge Univ. 48:219-248. Press, Cambridge. DARWIN, C. 1859. On the Origin of Species by ---. 1965. Evolutionary ecology.]. Anim. Eco!. Means of Natural Selection, or the Preservation 34:223-231. of Favoured Species in the Struggle for Life. LIDICKER, W. Z., JR. 1975. The role of dispersal Murray, London. in the demography of small mammals, p. 103­ GIFFORD, E. W. 1919. Field notes on the land birds 128. In K. Petrusewicz,. F. B. Golley and L. of the Galapagos Islands and of Cocos Island, Ryszkowski (eds.), Small Mammals: Their Pro­ Costa Rica. Proc. Calif. Acad. Sci. Ser. 4, 2:189­ ductivity and Population Dynamics. IBP, Vo!. 5. 258. Cambridge Univ. Press, Cambridge. GILPIN, M. E. 1975. Limit cycles in competition MAcARTHUR, R. H., AND E. O. WILSON. 1967. communities. Amer. Natur. 109:51-60. The Theory of Island Biogeography. Princeton GRANT, P. R. 1965. Plumage and the evolution of Univ. Press, Princeton. birds on islands. Syst. Zoo!' 14:47-52. RAup, D. M., AND S. ]. GOULD. 1974. Stochastic ---. 1966. Ecological incompatibility of bird simulation and evolution of morphology-to­ species on islands. Amer. Natur. 100:451-462. wards a nomothetic paleontology. Syst. Zoo!' ---. 1967. Bill length variability in birds of the 23:305-322. Tres Marias Islands, Mexico. Can. ]. Zoo!' RIGLER, F. H., AND R. R. LANGFORD. 1967. Con­ 45:805-815. generic occurrences of species of Diaptomus in ---. 1968. Bill size, body size and the ecological southern Ontario lakes. Can. ]. Zoo!. 45:81-90. adaptations of bird species to competitive situa­ SCHOENER, T. W. 1965. The evolution of bill size tions on islands. Syst. Zoo!' 17:319-333. differences among sympatric congeneric species ---. 1969. Colonization of islands by ecologi­ of birds. Evolution 19:189-213. cally dissimilar species of birds. Can. ]. Zoo!' ---. 1969. Size patterns in West Indian Anolis 47:41-43. lizards. 1. Size and species diversity. Syst. Zoo!' ---. 1970. Colonization of islands by ecologi­ 18:386-401. cally dissimilar species of mammals. Can. ]. SIMBERLOFF, D. 1969. Experimental zoogeography Zoo!. 48:545-553. of islands. A model for insular colonization. Ecol­ ---. 1972a. Convergent and divergent character ogy 50:296-314. displacement. Bio!. ]. Linnean Soc. 4:39-68. ---. 1970. Taxonomic diversity of island biotas. ---. 1972b. Interspecific competition among ro­ Evolution 24:23-47. dents. Ann. Rev. Eco!. Syst. 3:79-106. ---. 1976. Species turnover and equilibrium is­ ---. 1975. The classical case of character dis­ land biogeography. Science 194:572-578. placement. Evo!. Bio!. 8:237-337. ---. 1978. Using island biogeographic distri­ ---. 1980. The role of interspecific competition butions to determine if colonization is stochastic. in the adaptive radiation of Darwin's Finches. In Amer. Natur. 112:713-726. R. 1. Bowman and A. E. Leviton (eds.), Patterns SIMBERLOFF, D. S., AND E. O. WILSON. 1969. of Evolution in Galapagos Organisms. Amer. Experimental zoogeography of islands. The col­ Ass. Adv, Sci., Pacific Division, San Francisco. onization of empty islands. Ecology 50:278-296. HAIRSTON, N. G. 1964. Studies on the organization SNODGRASS, R. E. 1902. The relation of the food of animal communities. J. Anim. Eco!. to the size and shape of the bill in the Galapagos 33(Suppl):22 7-239. genus Geospiza. Auk 19:367-381. HASSELL, M. P., j. H. LAWTON, AND R. M. MAy. STRONG, D. R., JR., L. A. SZYSKA, AND D. S. SIM­ 1976. Patterns of dynamical behaviour in single­ BERLOFF. 1979. Tests of community-wide char­ species populations. J. Anim. Eco!. 45:471-486. acter displacement against null hypotheses. Evo­ HESPENHEIDE, H. A. 1973. Ecological inferences lution 33:897-913. from morphological data. Ann. Rev. Eco!. Syst, TERBORGH,]. 1971. Distribution on environmental 4:213-229. gradients: theory and a preliminary interpreta­ HUBBELL, S. P. 1979. Tree dispersion, abundance, tion of distributional patterns in the avifauna of and diversity in a tropical dry forest. Science the Cordillera Vilcambamba, Peru. Ecology 203:1299-1309. 52:23-40. INGER, R. F., AND R. K. COLWELL. 1977. Orga­ TERBORGH, ]., ]. FAABORG, AND H. ]. BROCK­ nization of contiguous communities of amphibi­ MAN. 1978. Island colonization by Lesser Antil­ ans and reptiles in Thailand. Eco!. Monogr. lean Birds. Auk 95:59-72. 47:229-253. WILSON, D. S. 1975. The adequacy of body size as KREBS, C. ]., AND]. H. MYERS. 1974. Population a niche difference. Amer. Natur. 109:769-784. cycles in small mammals. Adv. Eco!. Res. 8:267­ 399. Corresponding Editor: D. B. Wake LACK, D. 1945. The Galapagos finches (Geospizi-