Patterns of species co-occurrence and density compensation: a testfor interspecific competition in bat ectoparasite infracommunities

J. Sebastian Tello, Richard D. Stevens and Carl W. Dick

J. S. Tello (jtello1@lsu.edu) and R. D. Stevens, Dept of Biological Sciences, 107 Life Sciences Building, Louisiana State Univ., Baton Rouge, LA70803, USA. JST also at: Museo de Zoologıa, Escuela de Ciencias Biologicas, Pontificia Univ. Catolica del Ecuador, Av. 12 de Octubre 1076 yRoca, Apdo. 17-01-2184, Quito, Ecuador. � C. W. Dick, Dept of Zoology, Field Museum of Natural History, Chicago, IL 60605, USA.

Parasites constitute an ideal system with which to investigate patterns and mechanisms of community structure anddynamics. Nevertheless, despite their prevalence in natural systems, parasites have been examined less often than otherorganisms traditionally used for testing hypotheses of community assembly. In the present study, we investigate possibleeffects of competitive interactions on patterns of distribution (co-occurrence) and density among a group of streblid batflies parasitic on short-tailed fruit bats, Carollia perspicillata. Using null model analyses of species co-occurrence, we didnot find evidence that competition affects the distribution of bat fly species across hosts. Moreover, when non-infestedhosts were included, analyses showed evidence for interspecific aggregation, rather than for the segregation predicted bycompetition theory. Partial Pearson correlations among bat fly species densities showed no evidence of negativecovariation in two of three cases. In the species pair for which a significant negative correlation was found, a visualanalysis of plotted covariation indicated a constraint line, suggesting that competition between these two species mightbecome operational only in some infracommunities when abundances of bat flies approach a maximum set by one ormore limiting resources. Moreover, when a community-wide estimation of the significance of density compensation wascalculated, the result was not significant. Overall, we find no evidence that competition influences the distribution of batflies on their hosts, and mixed support for effects of competition on the densities of species. These results are consistentwith the idea that competition plays a role in structuring natural communities, but in many systems its effects are context-dependent and might not be important relative to other factors. Wider analyses across taxonomic and environmentalgradients and a detailed consideration of the different hypothesized effects of competition are necessary to fullyunderstand the importance of competition on natural communities.

Host�parasite systems offer a wealth of opportunity forunderstanding variation in distribution and abundance oforganisms (Gotelli and Rohde 2002). Most research hasfocused on documenting distribution and abundance ofparasites on hosts in an attempt to understand host�parasiteinteractions. Nonetheless, the relative frequency of inter-specific interactions among parasites, how these interactionsaffect the structure of such communities, and how this mayultimately influence host�parasite interactions remainsuncertain (Rohde 1991, Morand et al. 1999, Mouillotet al. 2005).

The nature of host�parasite relationships structuresparasite communities at several hierarchical scales, rangingfrom host clades and species to host individuals. Parasitecommunities assembled at this last scale are termed infra-communities (sensu Bush et al. 1997). The infracommunitylevel is the most fundamental of these scales of analysis sinceindividual hosts represent the units of distribution at whichmost local parasite interspecific interactions take place.

Parasite infracommunities make excellent model systemsfor understanding organization of natural communities.

Hosts typically provide discontinuous habitat patches thatcan support discrete communities of parasites (Gotelli andRohde 2002); this is in contrast to many terrestrial habitatsthat often subtly intergrade, thereby making delimitation ofcommunities difficult. Additionally, abundant hosts pro-vide ample and convenient replication with which toconduct powerful statistical analyses to assess hypothesizedprocesses underlying community structure (Holmes andPrice 1986). Individual hosts can also vary considerably interms of morphology, condition, fitness and life history,thereby providing natural treatment effects from which toexplore causative mechanisms.

Several mechanisms have been hypothesized to deter-mine variation in community composition and speciesabundance. Primary among these has been the classicalhypothesis of competitively induced deterministic structure.According to this hypothesis, species in a particularcommunity comprise a nonrandom subset of species avail-able from the regional pool, and possess characteristics thatminimize competitive interactions among them, therebyallowing coexistence (Hutchinson 1957, MacArthur and

Oikos 117: 693�702, 2008

doi: 10.1111/j.2008.0030-1299.16212.x,

# The Authors. Journal compilation # Oikos 2008

Subject Editor: Nicholas Gotelli. Accepted 28 January 2008

693

Levins 1967). The interspecific competition hypothesis hasbeen recently challenged from at least two differentperspectives: neutral theory and non-equilibrium ecology.The latter has been supported by some studies on parasitecommunities (Rohde 1991, Morand et al. 2002, Mouillotet al. 2003), suggesting that in these systems populationdensities often remain low, resources are not limiting, andconsequently communities do not become saturated. There-fore, neither resource-based competition nor zero-sumdynamics are likely to be important (Rohde 2005).However, other experimental and observational studieshave provided evidence that competition might be animportant force structuring parasite communities by caus-ing competitive exclusion (mainly among snail parasites,Kuris and Lafferty 1994) or by increasing differentiation ofmicrohabitat use within the host (mainly on platyhelmithworms, Bush and Holmes 1986). Studies on parasitecommunity structure, particularly those addressing theeffects of competition on composition and species abun-dance, have focused on a relatively limited set of parasitictaxa. Given the vast diversity of host�parasite systems, thegenerality of competitive interactions on parasite commu-nities remains unclear.

Ectoparasitic relationships between bat flies and theirchiropteran hosts provide an interesting and useful settingfrom which to better understand variation in communitystructure, as well as the mechanistic bases to such structure.Bat flies (Diptera: Streblidae) are obligate blood-feedingectoparasites exclusively associated with bats (Dick andPatterson 2006). Streblid bat flies exhibit high hostspecificity (Dick 2007, Dick and Patterson 2007), suggest-ing tight co-evolution between hosts and parasites in thissystem. Bat hosts are often abundant, especially in thetropics, and easy to sample (Rui and Graciolli 2005). Inparticular, the host species of this study, Carollia perspi-cillata, is widely distributed in South and Central Americaand is considered one of the most abundant species in theNeotropics (Cloutier and Thomas 1992). On a particularhost, bat flies can be numerous and are readily identifiable(Dick and Gettinger 2005). Hosts are often parasitized bymultiple fly species; in a survey in Venezuela, 63% ofinfested bats hosted simultaneously 2�4 fly species (Wenzel1976). Moreover, all bat flies consume a common resource(blood) and live in relatively small patches of habitat(hosts). How these ectosymbionts co-exist, given seeminglylimiting resources, remains a question.

In the present study, we test the hypothesis that inter-specific competition is a major force structuring infracom-munities of streblid flies that parasitize C. perspicillata. Weanalyze the distribution and densities of these parasitespecies to evaluate two predicted effects of competition onthe structure of natural communities. If competition is animportant force, it is expected that 1) competitors will bedistributed across communities in a way that avoids, or atleast reduces, their common occurrence (competitiveexclusion, less-than-expected patterns of species co-occur-rence), or that 2) in communities where competitors co-exist, abundance of one species will be inversely related tothe abundance of others (density compensation, negativecorrelations among the abundances of competing species).

Methods

Data collection

Individuals of Carollia perspicillata and their bat flyectoparasites were collected from several lowland andmountainside localities on both sides of the Andes inEcuador. Bat sampling at a particular site typically consistedof six mist-nets set at ground level within an area B1 km2

for up to five consecutive nights. When a C. perspicillatawas captured, it was immediately euthanized, removed fromthe net and placed into a sealed zip-lock plastic bag. At thefield laboratory, each bat and bag were carefully examinedfor streblids and other ectoparasites. This protocol mini-mized parasite escape and inter-host contamination, allow-ing for more accurate estimation of the proportion of hostsinfested (prevalence) and streblid individuals per parasitizedhost (intensity).

Bat flies were identified, individuals were quantified anda value of abundance (sensu Bush et al. 1997) was assignedfor each streblid species on each bat host examined. All batand bat fly specimens collected were deposited in thescientific collections of the Museo de Zoologıa of thePontificia Univ. Catolica del Ecuador (QCAZ).

Data analysis

Sampling was conducted at 18 localities. However, becausethe purpose of our study was to examine patterns of fly co-occurrence and abundance under the hypothesis of compe-titive interactions at the infracommunity level (on hostindividuals), we reduced the dataset to only those hostscollected at geographic localities from which all threestreblid species under study were recorded. By doing this,we intended to reduce or eliminate the effect of mechanismsacting at scales larger than the infracommunity, such asforces shaping geographic distribution or locality occupancyof the bat fly species. We acknowledge that one of theseforces could be competition, and that competitive interac-tions could have different effects at other scales of analyses(e.g. at the level of geographic locality or even region),which we are not addressing in this paper. The restriction oflocalities to those containing all bat fly species analyzed, andthe fact that sampling at each locality was homogeneous intime and space, maximized the probability that every hostcould be colonized by every bat fly species. This process ofdata filtering yielded information on the occurrence of 376streblids on 92 hosts collected at seven localities; fourof these localities were located west and three east ofthe Andes. The mean linear distance between localities was230 km.

Null model analyses of co-occurrenceGiven that resources are limiting, a superior competitorcould drive other competitors locally extinct (competitiveexclusion, Gause 1934). Competitively induced localextinction distributed across a landscape or a series ofhabitat patches would create a checker-board occupancypattern among competitors. Thus, competition theorypredicts less-than-expected patterns of species co-occurrence

694

(Diamond 1975). To test this prediction, null-models ofspecies co-occurrence were carried out.

Null models are statistical tools based on the randomiza-tion of empirical or theoretical data that allow comparisonbetween observed patterns and those expected under aparticular null hypothesis (Gotelli and Graves 1996). In thecase of null models of species co-occurrence, analyses areoften based on a presence�absence matrix. Columnsrepresent the units at which occurrences are measured(e.g. geographic localities, host individuals) and rowscorrespond to species. Each entry can be either 1 (presence),or 0 (absence). From this empirical matrix, a single co-occurrence index is extracted, representing the degree ofcommon occurrences of species throughout the commu-nities. Then, 1s and 0s are reshuffled in the matrix witheach occurrence being relocated independently of any other,but according to the constraints of a particular randomiza-tion algorithm. After a new random matrix has beencreated, the co-occurrence index is recalculated. Thisprocess is repeated a number of times, yielding a distribu-tion of the index under the null hypothesis of independenceamong occurrences. The empirical index can then becompared to the randomly generated distribution and avalue of its probability of outcome by chance can beobtained.

Delimitation of the empirical matrix is critical (Gotelliand Graves 1996). Thus, matrices must be logicallystructured and justified; but at the same time, they providea flexible means to modify the scenario in which hypothesesare tested and to further explore data. The decision toinclude or exclude empty sites is one that can havesubstantive effects on the outcome of randomizationanalyses. Some decisions are obvious, such as the exclusionof sites that fall outside of the dispersal capabilities of focaltaxa (Reddingius 1983). By limiting our samples togeographic locations where all species of ectoparasiteswere known to occur, we eliminated hosts that wereunsuitable in terms of isolation. Non-parasitized hostswithin the dispersal range of focal taxa present a quitedifferent scenario. Hosts could be empty because they arenot suitable habitats for ectoparasites, because by chancethey have not been colonized (stochasticity of assembly)(Gotelli and Rohde 2002), or because they represent spatialdynamics of parasite populations. Because of these diffe-rent interpretations, we analyzed our data including andexcluding non-parasitized bats. We agree with Gotelliand Rohde (2002) that complementary analyses with andwithout empty sites can provide important insights into theassembly of communities.

After our data were constrained to geographic localitieswhere all bat fly species were recorded, the remaining datawere used to build two matrices. The first consisted of 92columns, representing the total 92 hosts for which data onstreblid infracommunity structure were gathered. Thesecond matrix was the portion of the first that remainedafter removing all hosts that did not harbor any bat flyspecies; this process yielded a matrix with 75 columns.

Several indices have been proposed to quantify patternsof species co-occurrences, but not all are equally suitable fora particular dataset. In this study, the C-score was used.This index has the advantage that it does not require speciesto have perfect mutually exclusive distributions (Gotelli

2000), so it can be used in ‘‘noisy’’ datasets, where speciestend to competitively exclude each other, but still co-exist incertain communities. Gotelli (2000) showed that this‘‘noise’’, caused by imperfect divergent distributions ofspecies, can reach high values (around 50%) with C-scoresbeing still sufficiently powerful to detect less-than-expectedco-occurrences. Additionally, when coupled with appro-priate algorithms, it generates null models with goodstatistical behavior (Gotelli 2000). High values of thisindex indicate that the overall pattern is infrequent co-occurrence of species, while low C-score values mean thatspecies frequently occur together.

The final step in null model analyses of species co-occurrence is to select suitable randomization algorithms.Algorithms define constraints imposed on the process bywhich random matrices are generated. Gotelli (2000)demonstrated that algorithms maintaining the empiricalnumber of occurrences per species in the reshuffled matrices(fixed row algorithms) generally have good type I errorproperties (average false null hypothesis rejection B15%;Gotelli 2000). Thus, the null models used in this study werebased on four fixed-row algorithms, but with the followingconstraints imposed on the host (column) totals:

1. Fixed rows, equiprobable columns. All hosts have thesame probability of obtaining a parasite species duringthe randomization process.

2. Fixed rows, column probabilities given by host bodyweight. Considerable variation in host weight wasobserved (14.9�29.8 g). Host weight may be used as asurrogate for resource availability and habitat patchsize, which may in turn influence parasite speciesrichness per bat. Also, if body size is indicative of hostage, it can be associated to species richness by parasiteaccumulation (Pulkkinen and Valtonen 1999). Thesepotential effects were controlled using this algorithm,whereby probability of obtaining a parasite species isproportional to the host’s body weight.

3. Fixed rows, fixed columns. The total number ofspecies observed on each host is also maintained, andonly their identity is randomized.

4. Fixed rows, proportional columns. The probability ofa host acquiring a parasite species is proportional, butnot constrained to be identical, to the number ofparasite species found on that host in the empiricalmatrix.

The fixed-equiprobable algorithm differs from the others inthat it considers all hosts to be strictly homogeneous in thelikelihood of being colonized and the number of parasitespecies they can sustain. The other algorithms account forhost-to-host variation that could influence their suitabilityfor parasite colonization or coexistence. This host-to-hostheterogeneity can be known (body mass) or unknown, butassumed to be reflected in the number of parasite speciesobserved in the empirical data (in the case of fixed-fixed andfixed-proportional algorithms). Results stemming from thedifferent null model algorithms can indicate the degree thathost-to-host heterogeneity influences patterns of parasiteco-occurrence.

Only algorithms 1 and 2 were used in analyses thatincluded empty hosts because when occurrences are

695

reshuffled under constraints of fixed or proportionalcolumns (algorithms 3 and 4), the probability that anempirical empty host will receive a parasite species in any ofthe null matrices equals zero. On the other hand, all fouralgorithms were used in the scenario that excluded non-infested hosts. All null models were based on 10 000iterations and degenerate matrices were retained in all cases.Analyses were carried out using Ecosim 7.72 (Gotelli andEntsminger 2001).

Kuris and Lafferty (1994) have criticized the use ofobserved values of occupancy to parameterize co-occurrencenull models. They argue that if competition causesexclusion, then the empirical values of parasite occupancyare themselves already affected by competition. Conse-quently, their use could lead to detection of false randomdistributions, hiding real less-than-expected patterns of co-occurrence. This issue has the potential to influence ourresults since all algorithms constrained the null matrices tomatch the observed number of occurrences per species(fixed row algorithms). To address this concern, we re-analyzed our data following a solution advocated by Kurisand Lafferty (1994). Accordingly, occupancy values formodel parameterization must be calculated for each speciesusing only the fraction of the host sample that does not alsocontain a superior competitor. Although it is not possible toa priori define superior�inferior competitor hierarchiesamong our parasite species, we calculated values ofoccupancy for each bat fly species using only those hoststhat did not simultaneously harbor another parasite species.We then used the highest prevalence value for each bat flyspecies to parameterize our algorithms, and the null modelswere re-run. These results showed no more indication ofcompetitive exclusion than the original results, suggestingthat our analyses are not biased. Thus, we present resultsonly from the fixed row algorithms.

Pooling hosts from all localities into a single analysis mayobscure effects of spatial heterogeneity among geographiclocalities, thereby confounding our results. This possibilitywas suggested also by Kuris and Lafferty (1994) in theirstudy of larval trematode infracommunities. To account forthe effects of pooling sites, we re-analyzed our data bylocality and interpreted the overall trends by using a meta-analysis. The results and conclusions from these analyses(Supplementary material: Appendix 1) match closely thoseobtained from pooled data analyses, for which we developbelow.

Bi-variate correlations of species densitiesCompetitive exclusion, and the resulting less-than-expectedco-occurrence pattern, is a fundamental prediction made bycompetition theory. However, species may also respond tocompetition in other ways. For example, there can bechanges in abundance of a species depending on presence orabundance of competitors, observed primarily as negativedensity covariation (Houlahan et al. 2007); this response istermed density compensation (Stevens and Willig 2000).The observation of density compensation in natural systemssuggests that competition for a common limiting resource isdetermining abundances of species within the community.Thus, although co-occurrence patterns may themselves not

point to competition, patterns of abundance may still beindicative of competitive interactions.

Other factors, such as spatial heterogeneity of environ-mental variables, could also influence species abundancesand create spurious correlations. Thus, when testing fordensity compensation we controlled for environmentalheterogeneity by using partial Pearson correlations in whichvariation in locality elevation, host body weight and host sex(categorical dummy variable) was accounted for. Densitycorrelations were performed for each species pair using onlythose hosts that harbored one, the other, or both specieswhose densities were being considered (i.e. double-zerorecords were eliminated from all pair-wise correlations).Since we were exclusively interested in possible negativecorrelations, we used one-tailed probability values. Allpossible pair-wise combinations of species yielded threedifferent correlation analyses, and sequential Bonferronicorrections (Rice 1989) maintained alpha at 5%.

To obtain a community-wide estimation of the im-portance of density compensation in our study system, weused Fisher’s method to combine the probabilities of allthree independent correlations in a meta-analysis (Sokal andRolf 1995). This method consists of summing the naturallogarithms of the one-tail probabilities for each correlation,multiplying it by minus two, and comparing this value to ax2-distribution with 2k degrees of freedom (where k is thenumber of individual tests; Sokal and Rolf 1995).

Results

Four streblid bat fly species were recovered from Carolliaperspicillata: Trichobius joblingi (Fig. 1A), Strebla guajiro(Fig. 1B), Speiseria ambigua (Fig. 1C) and S. peytonae. S.peytonae was rare, represented by only two individuals in thetotal sample (B0.5% of the total number of bat fliescaptured). Also, this streblid species is known to beprimarily associated with a different, but closely related,host species: C. brevicauda (Wenzel 1976). Because of thisextremely low occurrence and the possibility that theserecords represent sampling contamination (Dick 2007), S.peytonae was omitted from all analyses. Trichobius joblingiwas the most common species (prevalence�0.7, meanabundance�3.11), followed by Strebla guajiro (pre-valence�0.35, mean abundance�0.50) and Speiseriaambigua (prevalence�0.27, mean abundance�0.48).

All null model algorithms exhibited qualitatively similarresults, indicating that host heterogeneity in body size andparasite species richness do not play important roles indetermining co-occurrence patterns. Nevertheless, differ-ences were observed between scenarios that excluded andincluded empty hosts. The analyses excluding non-infestedhosts showed that the observed co-occurrence pattern couldbe expected by simple random assembly (Fig. 2). However,the two null models that included empty hosts showed astrong positive non-random pattern of species co-occur-rence (Fig. 3), indicating that species co-occur more oftenthan expected by chance (Supplementary material: Appen-dix 1).

Of the three pair-wise correlations between speciesdensities, two showed no significant relationships (Fig.4A�B). In contrast, Speiseria ambigua and Strebla guajiro

696

exhibited a significant negative correlation (Fig. 4C).However, the pattern of density variation in these twospecies does not simply fall around a negative tendency line,but indicates a constraint. This constraint is graphicallyobserved as the triangular shape of the distribution ofvariation and the absence of points in the upper right cornerof the bi-variate plot (Fig. 4C). Finally, after combiningprobabilities from all three correlations, the overall sig-nificance of negative covariations among bat fly species wasnot statistically significant (x2�10.15, DF�6, p�0.118).

Discussion

The role of interspecific interactions in shaping patterns ofdistribution, assembly and abundance of species has beenbroadly investigated and remains a focus of ongoing debate.Much work has been conducted on free-living organisms,with less attention given to patterns of organization ofparasite communities and the mechanistic bases for theirstructure. In the present study, we investigated the role ofinterspecific competition in structuring simple bat ectopar-asite assemblages by testing two of its expected outcomes:competitive exclusion and density compensation.

Bat fly abundances and density compensation

Several studies that have investigated interspecific interac-tions through analysis of density correlation in parasiteshave reported both negative (Komeno and Linhares 1999,Alves and Luque 2001) and positive (Luque et al. 2003)relationships. However, lack of association also seems to becommon (Dean and Ricklefs 1979, Haukisalmi andHenttonen 1993, Behnke et al. 2005). In our examinationof density compensation, Fisher’s test for combiningprobabilities indicated overall non-significant negativecovariation, suggesting that competition may be relativelyunimportant in determining community-wide abundance

patterns of these species. Nevertheless, one species pair(Strebla guajiro and Speiseria ambigua) exhibited evidence ofdensity compensation, as expected by the effects of negativeinterspecific interactions. Species of the genera Strebla andSpeiseria have been reported to share similar microhabitatpreferences for furred body surfaces, while species in thegenus Trichobius tend to be located more often oninterfemoral and wing membranes (ter Hofstede et al.2004). These microhabitat preferences could set the stagefor the observed stronger interactions between Streblaguajiro and Speiseria ambigua.

Interestingly, visual inspection of the density covariationbetween these two species (Fig. 4C) reveals that points donot simply fall around a line of negative slope, but occurwithin an apparent upper constraint. It seems that whendensity of one species is low, density of the other species isfree to vary from zero to high abundance; but, when densityof one species is high, density of the other is constrained tobe low. This observation suggests that there are someinstances in which species densities become so high that oneor more resources turn into limiting factors and competi-tion becomes operational, setting a maximum to thenumber of individuals and species that can co-exist giventhe available resources. However, much variation in speciesabundances can occur below this constraint, and thisvariation may be attributable to other causes, such asdispersal limitation, disturbance or chance. In some situa-tions, mechanisms that prevent infracommunities frombecoming saturated may operate so frequently that speciesnever become abundant enough as to reach carryingcapacity and thus exhaust resources; in such cases, theseconstraints and competitive effects may never becomeapparent (Rohde 2005).

Parasite species distributions on host populations

Research on other groups of parasitic animals indicatesthat negative, random, or positive associations in the

Fig. 1. Relative body size and morphology of the three bat fly species that commonly co-occur on C. perspicillata: (A) Trichobius joblingi,(B) Strebla guajiro and (C) Speiseria ambigua.

697

distribution of coexisting species are all possible. Muchevidence for competitive exclusion among parasites comesfrom studies on larval trematodes on snail hosts (Kuris andLafferty 1994). However, a diverse set of study systemsshows random occurrence or interspecific aggregation,thereby bringing into question the importance of competi-tion in determining the assembly of parasites communities(e.g. helminths in voles, Haukisalmi and Henttonen 1993;Copepoda, Monogenea, Monopisthocotylea, Trematoda,Cestoda, Isopoda and Branchiura on marine fish, Gotelliand Rohde 2002; nematodes, cestodes and trematodes inantelopes, Fellis et al. 2003; nematodes, cestodes and

digeneans on mice, Behnke et al. 2005; fleas on rodents,Krasnov et al. 2006).

Our results support this line of evidence. When all hostswere incorporated into the analyses, the results indicate thatbat fly species are aggregated: they occur together moreoften than expected. This pattern is the opposite of thatpredicted by competition; thus, it represents clear evidenceagainst the idea that negative interspecific interactions drivethe pattern of distribution of parasite species across hostpopulations in our study system. However, such aggrega-tion can be caused by forces (below) that do not necessarilyinvolve positive species interactions. Consequently, there isstill the possibility that competition is important at asmaller scale of analyses, namely when only the part of thedistribution that represents reachable and suitable habitat isconsidered (parasitized hosts). In fact, it has been suggestedthat parasite aggregation could enhance conditions forinterspecific competition on the parasitized part of thehost population (Poulin 1998). Contrary to this idea, ourresults show that when empty hosts were excluded from theco-occurrence analyses, the emerging pattern was of randomassembly. Such a result is inconsistent with the competitionhypothesis, and strongly suggests that negative interspecificinteractions are not the main force affecting the distributionof parasites across habitat patches (hosts) in our studysystem. It also suggests that other forces may limit thedistribution of all species in a similar way, thereby causinginterspecific aggregation within host populations; but, thismechanism(s) may not operate (or be strong enough) toalso generate aggregation within the part of the hostpopulation that is parasitized.

When empty hosts were included in analyses, the fixed-equiprobable and the fixed-body weight algorithms did notprovide different results. This suggests that host differences

Fig. 2. Results of null model analyses based on the matrixexcluding empty hosts. The figure shows the position of theempirical C-score (arrow) in a frequency distribution of valuesgenerated randomly based on four different algorithms: (A) fixedrows�fixed columns, (B) fixed rows�proportional columns, (C)fixed rows�equiprobable columns, and (D) fixed rows�columnprobabilities given by host body weight. Note that box (A) doesnot share the scaling in the abscissa with (B), (C) and (D).

Fig. 3. Results of null model analyses based on the matrixincluding empty hosts. The figure shows the position of theempirical C-score (arrow) in a frequency distribution of valuesgenerated randomly based on two different algorithms: (A) fixedrows�equiprobable columns and (B) fixed rows�column probabil-ities given by host body weight.

698

in body mass do not explain the observed aggregation. Thisis further supported by a lack of correlation between hostbody weight and total parasite abundance, or parasitespecies richness in our system (r��0.044, p�0.706and r��0.001, p�0.992 respectively; unpubl.). Simi-larly, within the part of the host population that isparasitized, the similarity among results from the nullmodel algorithms suggests that the lack of structure is notassociated with host-to-host heterogeneity in body mass orprobability of colonization.

Mechanisms contributing to aggregationThe observed frequent co-occurrence when all hosts wereconsidered can result from positive species interaction, host-to-host (patch) heterogeneity in rates of dispersal andextinction, variability in host suitability for bat fly coloniza-tion or survival, or other evolutionary or ecologicalprocesses shaping distribution of the parasite species insimilar ways. First, positive indirect interaction amongparasite species have been suggested in which the presenceof a parasite species may reduce the effort of the host toavoid or eliminate individuals from other species. Facilita-tion of this nature would increase the chance for localcoexistence of various parasite species (Krasnov et al. 2005).This mechanism can be considered analogous to predator-mediated positive interactions (e.g. due to predator switch-ing) proposed to explain the coexistence of prey species infree-ranging organisms (Holt and Lawton 1994), where thepresence of a prey species enhances existence of another byreducing the net predatory effect. Predator-mediated coex-istence can be reached only if the effect of the predator isdependent on its own density, but its density is limited byother mechanisms besides prey exhaustion (Abrams andMatsuda 1996). In the case of parasite infracommunities,intensity of host anti-parasitic response (e.g. intensity ofgrooming behavior or immunological reaction, which isanalogous to predator density) must reach a limit setprimarily by constraints on the amount of energy or timethat can be invested in parasite avoidance or removal(Giorgi et al. 2001, Krasnov et al. 2005). In this way, theanti-parasitic response would increase relative to thenumber of parasites, until the threshold set by energeticconstrains is reached. Individuals or species numbersbeyond this point would be facilitated by a dilution ofthe anti-parasitic host effects.

Another possible explanation for interspecific aggrega-tion may relate to environmental and/or host characteristicsneeded by parasites to survive. This explanation requiresthat all parasite species have similar habitat preferences.This resemblance could be the case if there is a high degreeof conservatism in the evolution of niche characteristicsamong species (Weins and Graham 2005). However, thethree bat fly species belong to different genera, whichpresent substantial morphological differentiation (Fig. 1).In turn, this phenotypic dissimilarity is a sign of nicheevolution, divergence in niche characteristics and possiblyalso a lack of correspondence in host preference among batfly species. On the other hand, this apparent morphologicalsegregation could also suggest within-host microhabitatspecialization. Microhabitat partitioning has been found inmany other parasite systems (especially among Platyhel-

minthes; Bush and Holmes 1986, Cilso and Caira 1993),and could be an important way in which parasite speciesshare resources (Combes 2001), allowing their coexistence,or even allowing their aggregation (but see Rohde 1994 foran interpretation of parasite niche specialization unrelatedto competition).

The observed aggregation could also be related todispersal limitation. The role and importance of dispersalin patterns of community assembly and species abundancehas been studied and is now well supported (Leibold et al.2004). Since movement of individuals is strongly depen-dent on the spatial relationships among sites, dispersal couldbe especially important for parasites because their habitatpatches (hosts) are often not spatially static, but move andchange in position relative to each other. Thus, dependingon the host’s spatial interactions (e.g. movement andgrouping associated with social behavior), there can beheterogeneous rates of dispersal among them. Conse-quently, parasites may aggregate on those hosts with thehighest probabilities of transmission. This is supported bythe observation that hosts with higher chances of physicalencounter also have elevated rates of infestation (Arneberget al. 1998, Patterson et al. 2007). Particularly, in Carolliaperspicillata, sex-biased rates of contact among hosts canhelp explain the observed aggregation. The host species hasa polygynous mating system, where females form haremscontrolled by one dominant male; single males can formbachelor groups, but these are usually smaller and less stable(Cloutier and Thomas 1992). Consequently, female hostswould tend to accumulate more parasites. This possibleexplanation is supported by evidence suggesting that femaleC. perspicillata have higher loads of parasitism by bat flies(Fritz 1983, Tello and Jarrın unpubl.), which seems not tobe the result of a sex biased choice by the parasites (Dickand Dick 2006).

Finally, disturbance is also an important force shapingcommunity structure (Sousa 1984). Parasite infracommu-nities are likely subjected to near-continuous episodes ofdisturbance, because in most cases their habitat patches(hosts) are actively trying to get rid of them. Commonmanifestations of disturbance in parasite infracommunitiescan be host grooming behavior and host immunologicalresponses, which remove individuals from the system. Localextinction caused by the host could also exacerbate theeffects of dispersal limitation; if species are removed from asite, and there is a long lapse of time before dispersalrestores the species to that site, then species will tend toaggregate on the sites where local extinction due todisturbance is low or where re-colonization by dispersal ishigh. In vertebrates, heterogeneous rates of grooming andimmunocompetence can result from social behavior, sex,body size, age, etc. (Hart et al. 1992, Mooring et al. 1996,Roulin et al. 2003).

Concluding comments

Overall, we find no evidence of competition in patterns ofdistribution and mixed effects on the patterns of covariationin density among bat fly species. These results support theidea that, although there might be a signature in co-occurrence patterns of free-ranging species that is consistent

699

with competition theory (Gotelli and MacCabe 2002, butsee Ulrich 2004), it is often the case among parasites thatlimiting resources are not an important force shaping theassembly of their communities (Rohde 1991, 2005,Morand et al. 2002, Fellis et al. 2003, Krasnov et al.2005). On the other hand, our density compensation

analyses suggest that competition is present but notprevalent, being restricted to one pair of species. Moreover,the effects of competition might be important whenabundances of these species are high; but when a competitoris rare, the other species does not necessarily respond byincreasing in abundance. It is possible that in such lowabundance situations other factors become more importantdeterminants of species abundance. This is also consistentwith research on parasitic and free-ranging organisms thatshows that competition interacts with others mechanismsand is context dependent, being important in somesituations but unimportant in others (Combes 2001,Houlahan et al. 2007, Kaplan and Denno 2007).

Recent research on parasite community ecology hasprovided important insights into how parasite assemblagesare structured and how they might differ from moretraditional model organisms, but clear general trends havenot emerged. These studies have also indicated that it isnecessary to investigate how different hypothesized mechan-isms shape communities in a wider array of systems in orderto evaluate their effects across taxonomic and environmentalgradients. For example, some evidence suggests that thestrength with which competition might shape patterns ofco-occurrence and abundance could vary depending oncharacteristics of the taxon being analyzed (e.g. poiki-lotherms vs homeotherms, Gotelli and MacCabe 2002, orcore vs satellite species, Ulrich and Zalewski 2006). Thesedifferences in species and environmental characteristicsmight create communities that form a continuum rangingfrom random to deterministically structured (Rohde 2005,Fig. 11.1).

It is also important to understand and evaluate themultiplicity of effects that competition can have on naturalsystems. In this study, we have addressed the effects ofnegative species interactions based on limiting resources onthe co-occurrence and covariation in abundance of species,but other predictions from competition theory should alsobe tested. Competition can also be related to microhabitatspecialization (Reed et al. 2000, Friggens and Brown 2005)or patterns of morphology (e.g. morphological overdisper-sion, Stevens and Willig 1999, or character displacement,Dayan and Simberloff 2005). Habitat specialization hasbeen suggested among co-existing bat flies (ter Hofstedeet al. 2004) and phenotypic differences are also readilyapparent (Fig. 1). These potential differences in nicheutilization could be the reason for the lack of a strongercompetition signal in the distribution and abundance of batfly species in our study system. Evaluation of how ourresults match those of other bat fly species on differenthosts, locations and scales, and how this possible micro-habitat partitioning and morphological dissimilaritiesamong co-occurring bat flies fit predictions made by theinterspecific competition hypothesis or other possiblemechanisms of community assembly remains to be inves-tigated.

Acknowledgements � Thanks to Pablo Jarrın and Thomas Kunz(Boston Univ.) who provided funding and equipment to conductan important part of the fieldwork for this study. All three authorswere supported in part by grants from the National ScienceFoundation (JST and CWD by DBI-0545051; JSTV and RDS by

Fig. 4. Scatterplots of density for each of the three species pairsand partial correlation results. Sampling size for density correla-tions are different since for each pair-wise correlation only hostswith at least one of the two bat fly species being correlated wereincluded. See text for further explanation of the quantitativeanalyses.

700

DEB-0535939). Thanks to Mercedes Gavilanes, Juan PabloCarrera, Rene Fonseca, Carlos Carrion, David Lasso, MiguelPinto and Natalia Hernandez for help during fieldwork, and alsoto Dr. Ricardo Guerrero (Univ. Central de Venezuela) forassistance identifying a reference collection of bat flies. Thanksto Cristian Martinez (Field Museum of Natural History) forhelping produce Fig. 1. We are also grateful to Eve McCullochwho substantially improved the quality of our paper.

References

Alves, D. R. and Luque, J. L. 2001. Community ecology of themetazoan parasites of white croaker, Micropogonias furnieri(Osteichthyes: Sciaenidae), from the coastal zone of the stateof Rio de Janeiro, Brazil. � Mem. Inst. Oswaldo Cruz 96:145�153.

Abrams, P. A. and Matsuda, H. 1996. Positive indirect effectsbetween prey species that share predators. � Ecology 77:610�616.

Arneberg, P. et al. 1998. Host densities as determinants ofabundance in parasite communities. � Proc. R. Soc. Lond. B265: 1283�1289.

Behnke, J. M. et al. 2005. Do the helmith parasites of wood miceinteract? � J. Anim. Ecol. 74: 982�993.

Bush, A. O. and Holmes, J. C. 1986. Intestinal helminthes oflesser scaup ducks: an interactive community. � Can. J. Zool.64: 142�152.

Bush, A. O. et al. 1997. Parasitology meets ecology on its ownterms: Margolis et al. revisited. � J. Parasitol. 83: 575�583.

Cilso, P. R. and Caira, J. N. 1993. The parasite assemblage in thespiral intestine of the shark Mustelus canis. � J. Parasitol. 79:886�899.

Cloutier, D. and Thomas, D. W. 1992. Carollia perspicillata.� Mammal. Species 417: 1�9.

Combes, C. 2001. Parasitism. The ecology and evolution ofintimate interactions. � The Univ. of Chicago Press.

Dayan, T. and Simberloff, D. 2005. Ecological and community-wide character displacement: the next generation. � Ecol. Lett.8: 875�894.

Dean, J. M. and Ricklefs, R. E. 1979. Do parasites of Lepidopteralarvae compete for hosts? No! � Am. Nat. 133: 302�306.

Diamond, J. M. 1975. Assembly of species communities. � In:Cody, D. and Diamond, J. M. (eds), Ecology and evolution ofcommunities. Harvard Univ. Press, pp. 342�349.

Dick, C. W. 2007. High host specificity of obligate ectoparasites.� Ecol. Entomol. 32: 446�450.

Dick, C. W. and Gettinger, D. 2005. A faunal survey of streblidflies (Diptera: Streblidae) associated with bats in Paraguay. � J.Parasitol. 92: 1015�1024.

Dick, C. W. and Dick, S. C. 2006. Effects of prior infestation onhost choice of bat flies (Diptera: Streblidae). � J. Med.Entomol. 43: 433�436.

Dick, C. W. and Patterson, B. D. 2006. Bat flies: obligateectoparasites of bats. � In: Morand, S. et al. (eds), Micro-mammals and macroparasites: from evolutionary ecology tomanagement. Springer, pp. 179�194.

Dick, C. W. and Patterson, B. D. 2007. Against all odds:explaining high host specificity in dispersal-prone parasites.� Int. J. Parasitol. 37: 871�876.

Fellis, K. J. et al. 2003. Patterns of association, nestedness, andspecies co-occurrence of helminth parasites in the greaterkudu, Tragelaphus strepsiceros, in the Kruger National Park,South Africa, and Etosha National Park, Namibia. � J.Parasitol. 89: 899�907.

Friggens, M. M. and Brown, J. H. 2005. Niche partitioning in thecestode communities of two elasmobranchs. � Oikos 108: 76�84.

Fritz, G. N. 1983. Biology and ecology of bat flies (Diptera:Streblidae) on bats in the genus Carollia. � J. Med. Entomol.20: 1�10.

Gause, G. F. 1934. The struggle for existence. � Williams andWilkins, Baltimore. Reprint, p. 1971.

Giorgi, M. S. et al. 2001. The energetic grooming costs imposedby a parasitic mite (Spinturnix myoti) upon its bat host (Myotismyois). � Proc. R. Soc. Lond. B 268: 2071�2075.

Gotelli, N. J. 2000. Null model analysis of species co-occurrencepatterns. � Ecology 81: 2606�2621.

Gotelli, N. J. and Graves, G. R. 1996. Null models in ecology.� Smithsonian Inst. Press.

Gotelli, N. J. and Entsminger, G. L. 2001. EcoSim: null modelssoftware for ecology. Ver. 7.0, Acquired Intelligence Inc.& Kesey-Bear. Bhttp://homepages.together.net/�gentsmin/ecosim.htm�.

Gotelli, N. J. and MacCabe, D. J. 2002. Species co-occurrence: ameta-analysis of J. M. Diamond’s assembly rules model.� Ecology 83: 2091�2096.

Gotelli, N. J. and Rohde, K. 2002. Co-occurrence of ectoparasitesof marine fishes: a null model analysis. � Ecol. Lett. 5: 86�94.

Hart, B. L. et al. 1992. Biological basis of grooming behavior inantelope � the body size, vigilance and habitat principles.� Anim. Behav. 44: 615�631.

Haukisalmi, V. and Henttonen, H. 1993. Coexistence in thehelminths of the black vole Clethrionomys glareolus. I. Patternsof co-occurrence. � J. Anim. Ecol. 62: 221�229.

Holmes, J. C. and Price, P. W. 1986. Communities of parasites.� In: Anderson, D. J. and Kikkawa, J. (eds), Community

ecology: pattern and process. Blackwell, pp. 187�213.Holt, R. D. and Lawton, J. H. 1994. The ecological consequences

of shared natural enemies. � Annu. Rev. Ecol. Syst. 25: 495�520.

Houlahan, J. E. et al. 2007. Compensatory dynamics are rare innatural ecological communities. � Proc. Natl Acad. Sci. USA104: 3273�3277.

Hutchinson, G. E. 1957. Concluding remarks. � Cold SpringHarbor Symp. Quant. Biol. 22: 415�427.

Kaplan, I and Denno, R. F. 2007. Interspecific interactions inphytophagous insects revisited: a quantitative assessment ofcompetition theory. � Ecol. Lett. 10: 977�994.

Komeno, C. A. and Linhares, A. X. 1999. Batflies parasitic onsome phyllostomid bats in southeastern Brazil: parasitism ratesand host-parasite relationships. � Mem. Inst. Oswaldo Cruz94: 151�156.

Krasnov, B. R. et al. 2005. Covariance in species diversity andfacilitation among non-interactive parasite taxa: all against thehost. � Parasitology 131: 557�568.

Krasnov, B. R. et al. 2006. Are ectoparasite communitiesstructured? Species co-ocurrence, temporal variation and nullmodels. � J. Anim. Ecol. 75: 1330�1339.

Kuris, A. M. and Lafferty, K. D. 1994. Community structure:larval trematodes in snail hosts. � Annu. Rev. Ecol. Syst. 25:189�217.

Leibold, M. A. et al. 2004. The metacommunity concept: aframework for multi-scale community ecology. � Ecol. Lett. 7:601�613.

Luque, J. L. et al. 2003. Community ecology of metazoan parasitesof banded croaker, Paralonchurus brasiliensis (Osteichthyes:Sciaenidae), from the coastal zone of the State of Rio deJaneiro, Brazil. � Acta Sci. Biol. Sci. 25: 273�278.

MacArthur, R. H. and Levins, R. 1967. The limiting similarity,convergence, and divergence of coexisting species. � Am. Nat.101: 377�387.

Mooring, M. S. et al. 1996. Role of sex and breeding status ingrooming and total tick load of impala. � Behav. Ecol.Sociobiol. 39: 259�266.

701

Morand, S. et al. 1999. Aggregation and species coexistence ofectoparasites of marine fishes. � Int. J. Parasitol. 29: 663�672.

Morand, S. et al. 2002. Investigating patterns may reveal processes:evolutionary ecology of parasitic monogeneans. � Int. J.Parasitol. 32: 111�119.

Mouillot, D. et al. 2003. How parasites divide resources: a test ofthe niche apportionment hypothesis. � J. Anim. Ecol. 72:757�764.

Mouillot, D. et al. 2005. Parasite species coexistence and limitingsimilarity: a multiscale look at phylogenetic, functional andreproductive distances. � Oecologia 146: 269�278.

Patterson, B. D. et al. 2007. Roosting habits of bats affect theirparasitism by bat flies (Diptera: Streblidae). � J. Trop. Ecol.23: 177�189.

Poulin, R. 1998. Evolutionary ecology of parasites. � Chapmanand Hall.

Pulkkinen, K. and Valtonen, E. T. 1999. Accumulation ofplerocercoids of Triaenophorus crassus in the second inter-mediate host Coregonus lavaretus and their effect on growth ofthe host. � J. Fish Biol. 55: 115�126.

Reddingius, J. 1983. On species sharing islands. Comment on anarticle by S. J. Wright and C. C. Biehl. � Am. Nat. 122: 830�832.

Reed, D. L. et al. 2000. Spatial partitioning of host habitat bychewing lice of the genera Geomydoecus and Thomomydoe-cus (Phthiraptera: Thichodectidae). � J. Parasitol. 951�955.

Rice, W. R. 1989. Analyzing tables of statistical tests. � Evolution43: 223�225.

Rohde, K. 1991. Intra- and interspecific interactions in lowdensity populations in resource-rich habitats. � Oikos 60: 91�104.

Rohde, K. 1994. Niche restriction in parasites: proximate andultimate causes. � Parasitology 109: S69�S84.

Rohde, K. 2005. Nonequilibrium ecology. � Cambridge Univ.Press.

Roulin, A. et al. 2003. Which chick is tasty to parasites? Theimportance of host immunology vs parasite life history. � J.Anim. Ecol. 72: 75�81.

Rui, A. M. and Graciolli, G. 2005. Moscas ectoparasitas (Diptera,Streblidae) de morcegos (Chiroptera, Phyllostomidae) no suldo Brasil: associacoes hospedeiros-parasitos e taxas de infes-tacao. � Rev. Bras. Zool. 22: 438�445.

Sokal, R. R. and Rolf, F. J. 1995. Biometry. � W. H. Freeman andCompany.

Sousa, W. P. 1984. The role of disturbance in natural commu-nities. � Annu. Rev. Ecol. Syst. 15: 353�391.

Stevens, R. D. and Willig, M. R. 1999. Size assortment in NewWorld bat communities. � J. Mammal. 80: 644�658.

Stevens, R. D. and Willig, M. R. 2000. Density compensation inNew World bat communities. � Oikos 89: 367�377.

ter Hofstede, H. M. et al. 2004. Host and host-site specificity ofbat flies (Diptera: Streblidae and Nycteribiidae) on Neotropi-cal bats (Chiroptera). � Can. J. Zool. 82: 616�626.

Ulrich, W. 2004. Species co-occurrences and neutral models:reassessing J. M. Diamond’s assembly rules. � Oikos 107:603�604.

Ulrich, W. and Zalewski, M. 2006. Abundance and co-occurrencepatterns of core and satellite species of ground beetles on smalllake islands. � Oikos 114: 338�348.

Weins, J. J. and Graham, C. H. 2005. Niche conservatism:integrating evolution, ecology, and conservation biology.� Annu. Rev. Ecol. Syst. 36: 519�539.

Wenzel, R. L. 1976. The Streblid batflies of Venezuela (Diptera:Streblidae). � Brigham Young Univ. Sci. Bull. Biol. Ser. 20:1�177.

Supplementary material (available online as AppendixO16212 at Bwww.oikos.ekol.lu.se/appendix�).Appendix 1.

702

1

Oikos O16212Tello, J. S., Stevens, R. D. and Dick, C. W. 2008.Patterns of species co-occurrence and densitycompensation: a test for interspecific competition inbat ectoparasite infracommunities. – Oikos 000: 000–000.

Appendix 1. Null model analysis ofspecies co-occurrence by geographiclocalityPooling localities in our null model analyses could have biased ourresults by eliminating possible effects of spatial heterogeneity inthe observed levels of co-occurrence. This has the potential to hidepatterns of aggregation or segregation, which might be found if theanalyses were run at the geographic locality level. To assess thispossibility, we ran our null model analyses independently for eachlocality, and interpreted the results using a meta-analysis.

MethodsFor each locality, we ran the same null model analyses as we did forthe pooled data: we used four different null model algorithms andanalyses were carried out in two scenarios: including or excludingnon-infested hosts. However, we could not run all possible nullmodels in all localities. One of the localities incorporated in thepooled analyses was excluded from the single locality analyses be-cause it had only three hosts; in all other localities at least ninehosts were captured, and consequently had enough individuals torun the null models. In two localities 100% of the bats were para-sitized, consequently null models including empty host could notbe run. Also, the number of individuals was so small in two local-ities that the fixed-fixed algorithm could not randomize the ob-served matrix (this is caused because this algorithm sets many con-straints on how the observations must be reshuffled). In the end,we were able to run a total of 29 analyses by combinations of thescenario in which the algorithms were carried out (excluding orincluding empty hosts), the null model algorithm (fixed-equiprobable, fixed-fixed, fixed-proportional and fixed- by hostbody weight), and the locality (six localities).

From each of the previously described analyses, we were able tocalculate one-tailed probabilities for aggregation (p[Cobs≤Cexp])and for segregation (p[Cobs≥Cexp]), as well as a measure of standard-ized effect size

SES=(Cobs–Cexp)/Ssim

as used by Gotelli and Rhode 2002, where Cobs is the C-score cal-culated from the empirical matrix, Cexp is the mean C-score fromthe simulated matrices and Ssim is the standard deviation of the nulldistribution.

Exploring the effect of number of individuals on nullmodel analysis results

All null models used a total of 10 000 randomizations. However,the number of individuals that are included in each randomization

process could influence the results produced by the null modelsand the power of the tests. To understand these effects, we con-ducted linear regressions of the number of individuals included inthe randomization (n) on: 1) the standard deviation of the rand-omized distribution, 2) the standardized effect size, 3) and 4) bothone tail probability values. To control for non-independence of theobservations, we modified the number of degrees of freedom inthe denominator for the F-tests to be equal to the number of local-ities; this implied a reduction from 27 to 6 degrees of freedom. It isdifficult to determine exact degrees of freedom for the 29 pointsused in the regressions; however, we think that this reduction issufficient for the purpose of controlling for most non-independ-ence.

Testing standardized effect sizes against the zero nullhypothesis

To test whether the observed levels of co-occurrence differed sig-nificantly from random assembly, we compared the standardizedeffect size against the null hypothesis of no difference from zero.We did this by using a one-sample t-test for each combination ofalgorithm and scenario, and also for each algorithm irrespective ofthe scenario, and for each scenario irrespective of the algorithmused.

Testing for differences among algorithms andscenarios

To test if there are any trends or differences among the algorithmsor between the scenarios in which the null models were ran, weconducted three independent General linear models using as de-pendent variables: 1) the standardized effect size, 2) the probabilityof aggregation, and 3) the probability of segregation. In these anal-yses, algorithms and scenarios were included as independent varia-bles, while locality was included as a random factor.

ResultsIn Fig. A1, we observe that the number of individuals had a signif-icant impact on the null model outcomes. A greater number ofindividuals increases the possibility of more configurations for therandomized matrices, and consequently increases the variability inthe null distribution (Fig. A1a). Similarly, the number of individu-als has a significant association with the SES and the probabilityvalues (Fig. A1b–d). Increases in the number of individuals alsoincrease the power of the null model analysis (Fig. A1c–d), by in-creasing the SES (Fig. A1b).

Table A1 shows the results from the null model analyses runindependently per locality, algorithm and scenario. Most singlenull model analyses did not show a pattern that could be distin-

2

guished from random assembly. As suggested by the previous cor-relation analyses, this could result from a lack of statistical power.However, if we look at the locality that had the largest number ofindividuals (locality 3), we can see that all null model algorithmsthat excluded empty hosts were clearly not statistically significant,while the null models that included empty hosts show evidence ofinterspecific aggregation, being the only statistically significant re-sults at an alpha of 0.05. The results at this locality, where presum-ably we have enough statistical power, showed perfect concord-ance with the analyses that pooled all individuals together. On theother hand, at locality 5, three of the four null models that excludeempty hosts show marginally significant support for the idea ofsegregation.

Figure A2 and Table A2 show the results from the one-sample t-test of the standardized effect sizes with respect to the null expecta-tion of zero. Only one of the groups seems to be different fromzero: in the case of the null models that included empty hosts (irre-spectively of the null model algorithm), the one-sample t-testshowed marginally significant results.

Finally, the results of the General linear models can be found inTable A3 and Fig. A2 and A3. In general, the analyses that includeand exclude empty hosts seem to provide significantly differentresults, suggesting that when empty hosts are included, the levelsof co-occurrence are suggestive of patterns of aggregation, andwhen these hosts are not used in the analyses, the patterns seem tobe mostly random. This is supported by the statistical analyses

which show that the categories of scenario are significantly differ-ent for all three dependent variables. There also seems to be signif-icant variation in the results among the null model algorithms andlocalities, indicated by the significance of the algorithm term andthe interaction terms with locality. However, some of the observedinteractions with locality can result from differences in the statisti-cal power across sites which will show significant results in some,while no statistical support in others.

ConclusionsOverall, these results show support for our pooled-sites null modelanalyses. When the null models are run separately for each locality,there is little evidence of segregation patterns hidden in the analy-ses in which we pooled together hosts across localities. Moreover,when we conduct meta-analyses on the single locality results, theypoint to the same conclusions obtained from the pooled data inthat there is a clear difference between the results of the null mod-els when empty hosts are included versus when empty hosts areexcluded. It seems that when we consider the entire host popula-tion, these parasites are markedly aggregated; conversely, the para-sites seem to be distributed randomly in the part of the populationthat is infested. This suggests that competition is not a strong forceshaping the distribution and infracommunity assembly of theseparasite species.

3

Tabl

e A

1. R

esul

ts fr

om th

e sin

gle-

loca

lity

null

mod

el a

naly

ses.

F or e

ach

com

bina

tion

of L

ocal

ity, S

cena

rio

and

Alg

orith

m, f

or w

hich

we

ran

null

mod

el a

naly

ses o

f co-

occu

rren

ce, w

e re

port

the

tota

l num

ber o

f ind

ivid

uals

used

in th

e ra

ndom

izat

ion

(n),

the

obse

r ved

C-s

core

(Cob

s), th

e ex

pect

ed C

-sco

re (C

exp,

whi

ch is

the

mea

n of

the

C-s

cor e

s of t

he n

ull m

atri

ces)

, the

stan

dar d

dev

iatio

nof

the

null

dist

ribu

tion

(Ssim

), th

e on

e-ta

iled

prob

abili

ty fo

r agg

rega

tion

(p[C

obs≤

Cex

p]),

the

one-

taile

d pr

obab

ility

for s

egre

gatio

n (p

[Cob

s≥C

exp]

), an

d a

mea

sure

of s

tand

ardi

zed

size

effe

ct, a

s def

ined

by G

otel

li an

d R

hode

200

2 (S

ES)

. The

p-v

alue

s tha

t wer

e st

atist

ical

ly si

gnifi

cant

(p ≤

0.0

5) a

nd m

argi

nally

sign

ifica

nt (0

.05

≤ p

≤ 0.

10) a

re h

ighl

ight

ed in

bol

d le

tter

s.

Loca

lity

Scen

ario

Alg

orith

mn

Cob

sC

exp

S simp(

Cob

s≤C

exp)

p(C

obs≥

Cex

p)SE

S([

Cob

s–C

exp]

/Ssim

)

1E

xclu

ding

em

pty

host

sFi

xed-

ByH

ostW

eigh

t5

1.00

0.40

0.49

1.00

00.

401

1.22

11

Exc

ludi

ng e

mpt

y ho

sts

Fixe

d-Eq

uipr

obab

le5

1.00

0.39

0.49

1.00

00.

392

1.24

41

Exc

ludi

ng e

mpt

y ho

sts

Fixe

d-Pr

opor

tiona

l5

1.00

0.39

0.49

1.00

00.

387

1.25

71

Incl

udin

g em

pty

host

sFi

xed-

ByH

ostW

eigh

t9

1.00

3.10

1.56

0.17

20.

959

–1.3

431

Incl

udin

g em

pty

host

sFi

xed-

Equi

prob

able

91.

003.

071.

520.

168

0.96

0–1

.365

2E

xclu

ding

em

pty

host

sFi

xed-

ByH

ostW

eigh

t16

8.33

10.0

33.

170.

445

0.64

1–0

.537

2E

xclu

ding

em

pty

host

sFi

xed-

Equi

prob

able

168.

3310

.38

3.94

0.40

20.

679

–0.5

212

Exc

ludi

ng e

mpt

y ho

sts

Fixe

d-Fi

xed

168.

339.

070.

820.

410

1.00

0–0

.902

2E

xclu

ding

em

pty

host

sFi

xed-

Prop

ortio

nal

168.

338.

133.

880.

645

0.44

60.

053

3E

xclu

ding

em

pty

host

sFi

xed-

ByH

ostW

eigh

t25

12.6

712

.99

5.31

0.51

40.

528

–0.0

613

Exc

ludi

ng e

mpt

y ho

sts

Fixe

d-Eq

uipr

obab

le25

12.6

713

.11

5.29

0.50

50.

539

–0.0

843

Exc

ludi

ng e

mpt

y ho

sts

Fixe

d-Fi

xed

2512

.67

13.0

21.

200.

831

0.55

7–0

.298

3E

xclu

ding

em

pty

host

sFi

xed-

Prop

ortio

nal

2512

.67

11.1

35.

060.

657

0.38

90.

304

3In

clud

ing

empt

y ho

sts

Fixe

d-B

yHos

tWei

ght

3312

.67

37.1

611

.27

0.01

40.

990

–2.1

733

Incl

udin

g em

pty

host

sFi

xed-

Equi

prob

able

3312

.67

37.5

111

.41

0.01

30.

991

–2.1

774

Exc

ludi

ng e

mpt

y ho

sts

Fixe

d-B

yHos

tWei

ght

91.

672.

991.

660.

203

0.80

2–0

.798

4E

xclu

ding

em

pty

host

sFi

xed-

Equi

prob

able

91.

673.

101.

670.

187

0.81

9–0

.858

4E

xclu

ding

em

pty

host

sFi

xed-

Prop

ortio

nal

91.

672.

351.

620.

341

0.66

9–0

.423

5E

xclu

ding

em

pty

host

sFi

xed-

ByH

ostW

eigh

t7

5.33

2.98

1.40

0.97

30.

078

1.68

05

Exc

ludi

ng e

mpt

y ho

sts

Fixe

d-Eq

uipr

obab

le7

5.33

3.05

1.38

0.97

20.

085

1.64

65

Exc

ludi

ng e

mpt

y ho

sts

Fixe

d-Fi

xed

75.

335.

350.

390.

756

0.63

0–0

.052

5E

xclu

ding

em

pty

host

sFi

xed-

Prop

ortio

nal

75.

332.

811.

390.

980

0.06

31.

822

5In

clud

ing

empt

y ho

sts

Fixe

d-B

yHos

tWei

ght

85.

333.

541.

520.

919

0.17

71.

178

5In

clud

ing

empt

y ho

sts

Fixe

d-Eq

uipr

obab

le8

5.33

3.70

1.51

0.90

80.

203

1.07

66

Exc

ludi

ng e

mpt

y ho

sts

Fixe

d-B

yHos

tWei

ght

122.

002.

131.

780.

768

0.73

0–0

.073

6E

xclu

ding

em

pty

host

sFi

xed-

Equi

prob

able

122.

002.

171.

810.

762

0.73

6–0

.093

6E

xclu

ding

em

pty

host

sFi

xed-

Prop

ortio

nal

122.

001.

901.

750.

813

0.67

80.

055

6In

clud

ing

empt

y ho

sts

Fixe

d-B

yHos

tWei

ght

142.

006.

403.

170.

134

0.91

9–1

.388

6In

clud

ing

empt

y ho

sts

Fixe

d-Eq

uipr

obab

le14

2.00

6.47

3.12

0.12

70.

922

–1.4

31

4

Table A2. Results from one-sample t-tests to compare the values of standardized effect size to the null hypothesis of zero. See also Fig. A2.The p-values that were statistically significant (p ≤ 0.05) and marginally significant (0.05 ≤ p ≤ 0.10) are highlighted in bold letters.

Scenario Algorithm tstat DF p (two-tailed) Mean difference

Excluding empty hosts Fixed-Equiprobable 0.545 5 0.609 0.222Excluding empty hosts Fixed-Fixed –1.652 2 0.240 –0.418Excluding empty hosts Fixed-Proportional 1.474 5 0.200 0.511Excluding empty hosts Fixed-ByHostWeight 0.591 5 0.580 0.239Including empty hosts Fixed-Equiprobable –1.376 3 0.262 –0.974Including empty hosts Fixed-ByHostWeight –1.278 3 0.291 –0.931

All Fixed-Equiprobable –0.640 9 0.538 –0.256All Fixed-Fixed –1.652 2 0.240 –0.418All Fixed-Proportional 1.474 5 0.200 0.511All Fixed-ByHostWeight –0.570 9 0.583 –0.229

Excluding empty hosts All 1.132 20 0.271 0.218Including empty hosts All –2.026 7 0.082 –0.953

Table A3. Results from the GLM analyses for the standardized effect size (SES), the probability of aggregation (p[Cobs≤Cexp]) and theprobability of segregation (p[Cobs≥Cexp]). See also Figs. A2 and A3. The p-values that were statistically significant (p ≤ 0.05) andmarginally significant (0.05 < p ≤ 0.10) are highlighted in bold letters.

SES ([Cobs–Cexp]/Ssim) p(Cobs≤Cexp) p(Cobs≥Cexp)

F DF p F DF p F DF p

Scenario 13.268 1 0.036 9.497 1 0.054 9.334 1 0.055Algorithm 8.659 3 0.002 1.235 3 0.340 11.076 3 0.001Locality 3.365 5 0.151 2.261 5 0.246 4.021 5 0.102Scenario-Algorithm interaction 2.404 1 0.219 0.358 1 0.592 0.440 1 0.555Scenario- Locality interaction 2300.598 3 <0.001 8308.128 3 <0.001 1054.024 3 <0.001Algorithm- Locality interaction 242.049 12 <0.001 804.206 12 <0.001 218.883 12 <0.001

5

Fig. A1. Relationships among the number of individuals per locality (n) and (A) the variation in the null distribution of C-scores (Ssim),(B) the standardized effect size (SES), (C) the probability of aggregation (p[Cobs≤Cexp]), and (D) the probability of segregation(p[Cobs≥Cexp]). For each relationship the p-value has been adjusted for non-independence among points (see text for details). Algorithms:F-E: fixed-equiprobable, F-F: fixed-fixed, F-P: fixed-proportional, and F-W: fixed-probability by host weight.

6

Fig. A3. Comparison of: (A) probability of aggregation, and (B) probability of segregation among the categories of Scenario andAlgorithm. For each case: the box represent the 25 to 75 percentiles of the distribution, the bold horizontal line in the box represent themedian, and the top and bottom lines represent the maximum and minimum values respectively.

Fig. A2. Comparison of the standardized effect size values per Scenario and Algorithm categories. The dashed line across the figurerepresents the value expected under the null hypothesis of random assembly. For each case: the box represent the 25 to 75 percentiles ofthe distribution, the bold horizontal line in the box represent the median, and the top and bottom lines represent the maximum andminimum values respectively.