The effects of dispersal mode on the spatial distribution patterns of intertidal molluscs

Authors


M. Johnson, School of Biology and Biochemistry, The Queen’s University of Belfast, 97 Lisburn Road, Belfast, BT9 7BL, Northern Ireland, UK. Tel: + 028 90272297; fax: + 028 90236505; e-mail:m.johnson@qub.ac.uk

Summary

  • 1As many species of marine benthic invertebrates have a limited capacity for movement as adults, dispersal mode is often considered as a determinant of geographical ranges, genetic structure and evolutionary history. Species that reproduce without a larval stage can only disperse by floating or rafting. It is proposed that the colonization processes associated with such direct developing species result in spatial distributions that show relatively greater fine scale patchiness than the distributions of species with a larval dispersal stage. This hypothesis was tested by collecting molluscs at different spatial scales in the Isle of Man.
  • 2Spatial distribution patterns supported the predictions based on dispersal mode. Estimated variance components for species with larval dispersal suggested that the majority of the spatial variation was associated with variation between shores. In comparison, there was relatively more variability within shores for abundance counts of species with direct development.
  • 3Multivariate analyses reflected the univariate results. An assemblage of direct developers provided a better discrimination between sites (100 m separation) but the group of species with larval dispersal gave a clearer separation of shores (separated by several km).
  • 4The fine scale spatial structure of direct developing species was reflected in higher average species diversity within quadrats. Species richness also reflected dispersal mode, with a higher fraction of the regional species pool present for direct developers in comparison to species with larval dispersal. This may reflect the improved local persistence of taxa that avoid the larval dispersal stage.

Introduction

Variations in the perceived dispersal ability of benthic species are thought to have ecological and evolutionary consequences. Traits such as the presence and type of larvae have been correlated with the geographical range of species (Scheltema 1989; although see Bhaud 1993), the genetic structure of populations (Todd, Lambert & Thorpe 1998) and the persistence of species in the fossil record (Hansen 1978; Jablonski & Raup 1995). Positive relationships between the duration of larval development and geographical range are, however, contradicted by the paradoxically widespread distribution of certain molluscs that lack planktonic larvae (Johannesson 1988; O Foighil 1989). Such unexpectedly wide geographical distributions are thought to be the result of alternative dispersal mechanisms such as floating and rafting (Highsmith 1985).

There are a number of problems associated with using geographical range to examine the consequences of dispersal ability. These drawbacks include a reliance on secondary sources and that range size is a relatively crude measure of spatial distribution patterns. Sampling effort can vary between different secondary sources, introducing a variety of unknown biases (Scheltema 1989; Gaston 1996; Gaston & Blackburn 1999). Species identifications in different surveys may not be consistent. A recent review of marine invertebrates concluded that the available data were generally inadequate to test theories of rarity, including those involving questions of geographical range (Chapman 1999). The study described in this paper directly addresses the problems associated with using range sizes by collecting all data within the same survey and analysing the distribution data at different spatial scales.

An important drawback of comparisons between species based on geographical range information is that subtleties of abundance and spatial distribution are lost. These subtleties, however, form part of the hypothesis suggested to explain the widespread distribution of direct developing molluscs: Johannesson (1988) proposed that colonization success would be higher for direct developers such as Littorina saxatilis (Olivi) in comparison to Littorina littorea (L.), which has a planktonic dispersal stage. The higher colonization success of direct developers was thought to reflect the relatively greater reproductive success of L. saxatilis founder groups as a result of the retention of offspring within a restricted area (Johannesson 1988). If this is a general mechanism, the distributions of direct developers should tend to show more fine scale structure than those of planktonic developers. Under this model, the spatial pattern of direct developers should reflect the persistence of small patches formed by stochastic colonization events. The patches are predicted to be relatively small in comparison to the shore area as each colonization event originates from a discrete group of rafting individuals (e.g. Ingolfsson 1995). Direct developers should also be less sensitive than planktonic developers to variability in hydrographic conditions, as established populations of brooding species are not dependent on the supply of larvae from other locations. In comparison to the situation with direct developers, larval supply is often emphasized as an important determinant of population abundance in species with a planktonic stage (e.g. Gaines, Brown & Roughgarden 1985; Minchinton & Scheibling 1991; Gaines & Bertness 1992). Variations in larval supply may be reflected in population variability at a range of spatial scales from less than a metre to several kilometres (Lindegarth, Andre & Jonsson 1995; Underwood & Chapman 1996; Hyder et al. 1998). Shoreline configuration alters local hydrodynamic patterns; this can add to the large-scale component of spatial variability in larval abundances and recruitment (Archambault & Bourget 1999). Hence benthic species with larval stages seem likely to show relatively more variability at larger spatial scales than direct developing species.

Information on the spatial scales of variability in species density can be summarized using nested analysis of variance (anova, Morrisey et al. 1992; Chapman 1994). The relative contribution of each spatial scale to variability in abundances can be calculated from the variance component for each level in the anova (Sokal & Rohlf 1995). Clearly the spatial distribution of a species can be affected by factors other than dispersal mode. These include habitat preferences, biotic interactions such as competition and predation, and abiotic disturbances. However, it seems unlikely that effects relating to other impacts on distribution could reproduce the predictions relating to dispersal mode. Considering gastropods of approximately the same size, there seems to be no obvious mechanism that would bias the effects of predation on spatial distributions to mimic the predictions based on dispersal mode. Of course, processes such as disturbance or predation will still affect spatial pattern. The null hypothesis is therefore that the various influences on spatial distribution obscure any effects related to dispersal mode, leading to no consistent patterns of population variability with spatial scale. In contrast, the hypothesis elaborated from the ideas of Johannesson (1988) is that spatial variability will be concentrated at smaller (within shore) scales for direct developing species, with relatively more variability at between shore scales for species with planktonic larvae.

Relationships between dispersal mode and the relative sizes of population variability at different spatial scales should also be reflected by patterns of community structure. By looking at species assemblages it is possible to include information on species that are too infrequently collected to be used in a full anova design. This multivariate approach leads to the prediction that assemblages consisting of direct developing species should show more variability in species composition at small spatial scales and less variability between larger spatial scales than assemblages composed of species with planktonic dispersal.

Materials and methods

Sampling sites

Samples were collected using a hierarchical design from shores in the Isle of Man during July 1999 (Fig. 1). The largest scale in the design involved a comparison between east and west coasts of the Island. Surveys of Littorina littorea imply that densities are higher on the west coast than on the east coast, possibly reflecting differences in larval supply (Norton et al. 1990). A potential hydrographic cause of differences in larval supply relates to the presence of a frontal region adjacent to the west coast of the Isle of Man (Simpson & Hunter 1974; Fig. 2 of Fogg et al. 1985). Increased concentrations of mollusc larvae have been recorded in Irish Sea fronts (Scrope-Howe & Jones 1985). Shoreward movement of the front may therefore cause increased recruitment to intertidal communities (Pineda 1994; Shanks et al. 2000). The persistence of species with direct development should be independent of hydrographic factors related to larval supply. Species with larval development are therefore predicted to show greater differences in abundance between east and west coasts than species with direct development. Smaller-scale sampling consisted of two shores selected randomly from those available on each coast with two sites within each shore. Sites were separated by approximately 100 m and samples were taken from six haphazardly thrown 0·25 m2 quadrats at each site. General texts imply greater molluscan species richness at locations lower on the shore (Fretter & Graham 1962). Samples were therefore taken just above the level of low water on spring tides to maximize the number of species available for contrasts.

Figure 1.

Location of the Isle of Man in the central Irish Sea (inset) and locations of shores sampled.

Figure 2.

Relationship between the mean and variance of individuals counted in quadrats (n = 48). Each point represents the summary statistics for a different species. ▴, Direct development; ▪, planktonic dispersal; ●, dispersal mode undefined; —, ln(variance) = 1·898 + 1·553ln (mean) r2= 96%, P < 0·001.

Collection and analysis of molluscs

All mollusc species present in sample quadrats were removed, along with any algae present. The algal material collected in quadrats was carefully washed in freshwater using a 1-mm sieve to collect individual molluscs. Samples were fixed in formalin (4% in seawater) and transferred into industrial methylated sprits (74%) after four days. Collected material was identified using Tebble (1966), Jones & Baxter (1987), Graham (1988) and Thompson (1988). Species authorities are given in Table 1. Littorina neglecta Bean has a controversial status, with some authorities regarding it as an ecotype of Littorina saxatilis (Olivi) (Reid 1996) and others maintaining a species status (Hull, Grahame & Mill 1999). While retaining a species rank for L. neglecta, we note that both taxa were infrequent in our samples.

Table 1.  Taxa collected from the Isle of Man sorted by frequency of occurrence (% frequency in quadrats)
Taxa with direct developmentFrequency in quadratsTaxa with larval dispersalFrequency in quadrats
Lacuna parva (da Costa 1778)85Mytilus edulis Linnaeus 175877
Onoba semicostata (Montagu 1803)69Helcion pellucidum (Linnaeus 1758)71
Littorina fabalis (Turton 1825)60Rissoa parva (da Costa 1778)54
Turtonia minuta (Fabricius 1780)50Gibbula cineraria (Linnaeus 1758)44
Musculus discors (Linnaeus 1767)44Patella vulgata Linnaeus 175833
Nucella lapillus (Linnaeus 1758)37Anomiacea indet.23
Littorina obtusata (Linnaeus 1758)23Venerupis senegalensis (Gmelin 1791)21
Skeneopsis planorbis (Fabricius 1780)23Gibbula umbilicalis (da Costa 1778)17
Cingula cingillus (Montagu 1803)23Lacuna vincta (Montagu 1803)15
Rissoella diaphana (Alder 1848)12Hiatella arctica (Linnaeus 1767)12
Calliostoma ziziphinum (Linnaeus 1758)10Tectura virginea (Müller 1776)10
Limapontia senestra (Quatrefages 1844) 6Patella aspera (Röding 1798)10
Littorina neglecta Bean in Thorpe 1844 2Tonicella rubra (Linnaeus 1767) 8
Littorina saxatilis (Olivi 1792) 2Tricolia pullus (Linnaeus 1758) 8
  Odostomia unidentata (Montagu 1803) 8
Dispersal mode undetermined Retusa truncatula (Bruguiére 1792) 4
Rissoidae indet. 4Lepidochitona cinerea (Linnaeus 1767) 2
Nudibranchia indet. 2Acanthochitona crinita (Pennant 1777) 2
Galeommatacea indet. 2Gibbula magus (Linnaeus 1758) 2
Tellinacea indet. 2Alvania semistriata (Montagu 1808) 2
  Kellia suborbicularis (Montagu 1803) 2

Details of reproductive form were taken from Fretter & Graham (1962) for most prosobranchs except Lacuna parva (da Costa) (Ockelmann & Nielsen 1981). Classification into direct developers or larval dispersers follows Quayle (1952), Ockelman (1964), Bayne (1976), Mackie (1984) and Maximovich (1984) for bivalves; Thompson (1976) for opisthrobranchs and Jones & Baxter (1987) for chitons. Processed samples were catalogued and stored at the National Museum of Scotland (accession number NMSZ1999254). The shore at Niarbyl had a much higher density of small Mytilus edulis L. than the other shores surveyed. Mats of small mussels are difficult to remove without destroying specimens. Mytilus counts from Niarbyl therefore underestimated densities and were not comparable to counts from the other shores. As the data on mussels were biased, they were excluded from the multivariate tests and the univariate comparisons between direct developers and species with planktonic larvae. Results of nested anova for M. edulis are, however, given for comparison with the other species.

Untransformed count data typically showed evidence of heterogeneity of variances, making the raw data unsuitable for anova. Relationships between the variance and mean of quadrat counts were therefore used to select a suitable transformation before analyses using anova. If changes in variance with increasing mean values are consistent, Taylor’s power law can be used to provide an appropriate transformation based on the slope of log-transformed plots (Legendre & Legendre 1998). Subsequent analyses using transformed data were checked for homogeneity of variances using Cochran’s test (Underwood 1997).

Multivariate comparisons were based on a division of the count data into separate community matrices: one matrix representing all species with direct development and the other matrix composed of species with planktonic dispersal stages. Each matrix was composed of transformed counts (x0·224, see Results section and note similarity to x0·25 transformation commonly used in community analyses; Clarke & Warwick 1994). Rows in each matrix represented species, with separate columns for each quadrat. The similarity between the species counts of each pair of quadrats (i and l) was calculated using the Bray–Curtis coefficient, Sil:

image( eqn 1)

where yij and ylj are the counts for species j in samples i and l, respectively, and n is the total number of species in each community matrix. Identical sample counts in a pair of quadrats result in a Bray–Curtis coefficient of 100. The coefficient is zero if there are no species in common between separate quadrats. Patterns in the similarity between groups of quadrats can be examined using the anosim (analysis of similarities) subroutine of the primer statistics package (Clarke & Green 1988; Clarke & Warwick 1994). The anosim procedure is based on calculating a test statistic that compares the average of rank similarities between different predefined groups with the average rank similarities within the same groups. If the predefined groups represent a genuine source of structure within the community matrix, the test statistic will be positive, with a maximum value of 1. The statistical significance of the test statistic of dispersion is estimated using a randomization process. The anosim procedure can also be applied to simple nested designs consisting of predefined groups and subgroups (Clarke & Warwick 1994).

Interactions between taxa or habitat preferences may influence the counts in quadrats. For example, two species may be associated with a particular habitat type, such that spatial variability in counts tracks the availability of the habitat type. These kinds of links between counts of different taxa could under or over-emphasize any potential effects of dispersal mode. Hence, the interdependence of counts was examined using all possible cross-taxa correlations.

Results

A total of 8580 molluscs, belonging to 39 taxa, were identified in samples (Table 1). Dispersal mode did not seem to be confounded with other factors such as life span, diet or feeding mode. For example, annual species such as Lacuna spp. and longer-lived species (e.g. Patella spp., Nucella lapillus (L.)) were found in both direct developing and larval dispersal groups. Species counts typically showed evidence of clumped distributions, with variance to mean ratios exceeding one. The relationship between the variances and means of different species did however, show consistent scaling (Fig. 2). More abundant species tended to show increasingly clumped distributions, indicated by a regression slope greater than 1. There were no clear trends in the sign or magnitude of residuals between different dispersal modes. The regression slope of the log-transformed plot suggested an x0·224 transformation for subsequent analyses of count data.

Analyses using nested anova were restricted to taxa that occurred on each of the four shores. Heterogeneity of variances in less common taxa could not be removed by transformation due to the presence of large numbers of zero counts. Significant variation occurred at different spatial scales in analyses of variance using the most common species (Table 2). There was little variation in abundances from one coast to the other. In seven out of 10 cases there appeared to be no additional variation associated with a comparison between coasts relative to the differences between shores (F ratios below 1). Heterogeneous variances increase the probability of detecting an effect where none exists (Type I error). Therefore an alternative (square root) transform was applied to counts of Rissoa parva (da Costa) to investigate the results. This alternative transform removed heterogeneity of variances and the significant differences between coasts, suggesting that the differences between east and west coasts in the original analysis for R. parva were due to Type I error. The incomplete information on spatial variability of Mytilus edulis also indicated that there was no significant variation between coasts. The most common class of significant difference in the nested anova was differences between shores. However, ranks of separate shores by abundance were not consistent among different taxa. There was always a degree of additional variation associated with differences between sites for species with direct development (F ratios exceeding 1). In comparison, for the majority of planktonic dispersers, there was no additional variation associated with differences between sites. A Fisher’s exact test of the proportion of taxa where F ratios exceeded 1 at the site scale indicated a significant association with dispersal mode (P < 0·05).

Table 2.  Summary of nested analyses of variance for the most abundant mollusc taxa. Variances of raw data were generally heterogeneous and all analyses used x0·224 transformed data. Probability levels are given as: *** P < 0·001, ** P < 0·01, * P < 0·05, NS = not significant
SpeciesSource of variation
Coast (d.f. 1)Shore (d.f. 2)Site (d.f. 4)Residual (d.f. 40)
MSFPMSFPMSFPMSCochran’s
(a) Direct
     L. parva 4·54 0·91NS5·00 9·04*0·551·10NS0·50NS
     L. fabalis 0·00 0·00NS9·9416·47*0·601·84NS0·33NS
     M. discors 0·34 0·21NS1·61 4·69NS0·341·38NS0·25NS
     N. lapillus 0·68 0·67NS1·01 1·93NS0·532·05NS0·26NS
     O. semicostata 0·01 0·00NS4·20 3·42NS1·233·78*0·32NS
     T. minuta 7·25 4·06NS1·79 4·84NS0·371·06NS0·35NS
(b) Planktonic
     G. cineraria 0·67 0·09NS7·57 9·83*0·772·61*0·30NS
     H. pellucidum 1·52 0·25NS6·1446·4**0·130·64NS0·21NS
     P. vulgata 0·66 7·48NS0·09 0·44NS0·200·64NS0·31NS
     R. parva19·9024·31*0·82 3·63NS0·220·83NS0·27*

The hypothesis of relatively greater spatial variability at within shore scales for direct developing species compared to relatively more variability at between shore scales for species with planktonic larvae was tested with a simple anova model: partitioning the variation into within-shore or between-shore scale and above. Results from the full nested model (Table 2) support this simplification as most significant differences were at the shore scale. The relatively weak effects associated with differences between coasts are unlikely to confound a simpler model that ignores variation at this upper spatial scale. The relative sizes of variance components from a one-way anova for shores are shown in Table 3. Components for Patella vulgata L. were not resolved as the mean square for between shore variation was less than the within shores mean square, contradicting the results of the full nested model (Table 2). The percentage of the total variance associated with within shore sources of variability was significantly higher for the group of direct developing species than the group of species with planktonic dispersal (t-test calculated for unequal variances, t = 3·49, P < 0·05). For all three species with planktonic dispersal, the majority of the spatial variability in counts was associated with variation between shores. Analysis of the available data for Mytilus edulis also indicated that the majority of the variance in counts (76%) was associated with differences between shores.

Table 3.  Variance component estimates from a simplified anova model. Data transformed (x0·224) before analysis. Cochran’s test implied homogeneity of variances in all cases (P > 0·05)
SpeciesBetween shores (%)Within shores (%)
(a) direct
     L. parva41·5858·42
     L. fabalis59·6840·32
     M. discors23·1176·89
     N. lapillus15·5884·42
     O. semicostata32·9367·07
     T. minuta43·5756·43
(b) planktonic
     G. cineraria54·8345·17
     H. pellucidum64·5335·47
     P. vulgata
     R. parva68·2731·73

The results of multivariate analyses of species assemblages reflect the anova results (Table 4). Sites within shores could be discriminated using an assemblage of direct developing species, but not on the basis of species with planktonic dispersal. The community compositions of separate shores were significantly different, regardless of dispersal mode. The community matrix based on species with planktonic dispersal, however, had a higher anosim test statistic value. Hence species assemblages were more variable between sites within shores for direct developers but more variable between shores for planktonic dispersers.

Table 4.  Multivariate dispersion indices calculated from the nested ANOSIM procedure carried out separately on all species with direct development and on all species with a planktonic stage. Probability levels are given as: *** P < 0·001, ** P < 0·01, * P < 0·05, NS = not significant
 Between shoresPBetween sitesP
Direct developers0·542*0·197**
Planktonic dispersers0·896**0·078NS

Relationships between counts of different taxa did not appear to influence the results. Cross-correlations suggested that taxon counts were statistically independent of each other. The average cross-correlation between all possible combinations of taxa was 0·035 (SD 0·184). Correlations within and between dispersal modes were also weak. The average cross-correlation between the most abundant taxa with direct development was 0·146 (SD 0·253). The average cross-correlation between pairs of taxa with different dispersal modes was 0·083 (SD 0·319). There were slightly stronger negative correlations within the group of abundant taxa with planktonic dispersal. However, the mean cross-correlation between taxa with planktonic dispersal was still near zero (−0·107, SD 0·218).

Discussion

The results clearly indicate that dispersal mode can affect the spatial distributions of intertidal molluscs as hypothesized. The potential effects of habitat preferences, biotic interactions and abiotic disturbances did not obscure the different spatial patterns associated with direct development or planktonic larvae. Possession of a planktonic larval stage compels there to be some dispersal between successive generations. This leads to relatively more variability at larger spatial scales (between shores) than is the case for species with direct development. The finer structure of within shore spatial variation for taxa with direct development was reflected in multivariate analyses. An assemblage of direct developers provided a means of discriminating between sites within shores whereas an assemblage of species with larval dispersal did not. Although spatial patterns of molluscs with direct development or planktonic larvae do not seem to have been compared before, previous studies report the influence of larval duration on spatial patterns of abundance. Species with short-lived planktonic propagules may have more patchy distributions and restricted colonization dynamics in comparison to species with longer-lived planktotrophic larvae (Bingham 1992; Reed et al. 2000).

The differences between the spatial patterns of direct developers and larval dispersers may be the result of responses to within-shore disturbances. Under this model it takes longer for direct developers to recolonize after removal of the populations at a disturbance site (Reed et al. 2000). This implies that abundances of different direct developer species will be correlated due to repeated low abundances at disturbance sites. The observed cross correlations between direct developers were, however, weak. This suggests that random colonization events play a greater role than within shore disturbances in determining the observed spatial patterns.

It was not possible to link the spatial patterns of variability with large scale hydrographic features as there was generally no detectable variation between separate coasts. It is possible that generalities about the effects of fronts on recruitment, particularly where the oceanographic mechanisms are different, do not apply to the system on the west coast of the Isle of Man. Without direct measurements, however, the effects of variations in larval supply and postrecruitment mortality cannot be separated.
Thorson (1950) proposed that the variable influences of hydrography and biotic interactions on pelagic eggs and larvae are reflected by more temporal variation in adult stocks when compared to species with limited or no planktonic reproductive stages. If planktonic species are more variable, this suggests that recruitment of larvae is the principal determinant of adult population size. Later authors have challenged Thorson’s generalizations, emphasizing the importance of post-recruitment processes such as competition, predation and physical disturbance (Levin & Hugget 1990; Olafsson, Peterson & Ambrose 1994; Menge 2000). This study does not contribute directly to the debate about the relative importance of pre- and post-recruitment regulation of population sizes. However, the results reflect the importance of reproductive form as proposed by Thorson (1950), with the emphasis on spatial pattern rather than population dynamics. The spatial signature of different dispersal modes was retained across a diverse range of molluscan species including grazers, filter feeders and predators despite the disparate post-recruitment processes that must have been occurring.

The different patterns of spatial variation in direct developers and species with planktonic larvae represent interacting scales of patchiness and have potential consequences for competitive interactions, predation and disease (Atkinson & Shorrocks 1981; Whitlatch et al. 1997; Doak 2000; Fauchald, Erikstad & Skarsfjord 2000). For example, coexistence between a superior and an inferior competitor is possible when the species have independently clumped distributions (Atkinson & Shorrocks 1981). Further speculation about the consequences of differences in spatial variability is, however, limited by the lack of ecological information about many of the species collected. For example, little is known about the strengths of competitive interactions between the different species.

Patterns of species diversity appear to reflect the increased small-scale variability of direct developers in comparison to larval dispersers. Within-quadrat diversity for taxa with direct development was significantly higher than for the group of species with planktonic larvae (t-test, P < 0·05, average Shannon–Weaver index (H′) for direct developers = 0·41 (SE 0·02) compared to 0·31 (SE 0·03) for species with planktonic larval dispersal). These differences were due to an increased evenness in counts as there were fewer species with direct development collected. When comparing patterns of species richness, the fraction of the total species pool in samples was always higher for direct developers, even when aggregating samples at different scales (Fig. 3). Both scale of observation and dispersal mode effects were significant (F ratios of 19·4 and 27·1, respectively, P-values less than 0·001, interaction not significant) in a two-way anova using random subsampling of quadrat and site data to produce a balanced design. It seems likely that the differences in species richness between dispersal modes reflect the persistence of species in a particular location following colonization. For species with larval dispersal, persistence is dependent on recruitment from the plankton each generation. The species present in any location therefore represent recent colonists. In contrast, local populations of species with direct development may persist by producing juveniles in situ. Relatively higher numbers of direct developing species may therefore represent an accumulation of different colonizations over successive generations.

Figure 3.

Average fraction of the total species pools of direct developers or larval dispersers present in samples aggregated by quadrat, site or shore. Error bars are ± SE. ●, Direct development; ○, larval dispersal.

It is not clear to what extent the patterns associated with different dispersal modes can be applied to other habitats. Individuals living in soft sediment habitats can be redistributed by post-settlement sediment transport (Hewitt et al. 1997). Habitat contrasts also occur within taxa. Population changes of a direct developing gastropod reflect floating juveniles on sandy shores but crawling adults on rocky shores (Adachi & Wada 1999). Extensive redistribution of individuals by sediment transport or floating may obscure the finer scale patterns predicted for species with direct development.

The analyses presented here emphasize alternatives to geographical range when comparing the spatial distributions of species. The nested approach allowed testing of more subtle hypotheses that those associated with range size and avoided the biases associated with extrapolating from secondary sources. Further work is needed to describe the generality of the observed patterns in space and time. Definition of the scales of interest is essential to most hypotheses about population variability (Ray & Hastings 1996). To re-examine Thorson’s hypothesis about the relative stability of populations with different modes of reproduction, the appropriate scale is that of the smallest coherent patch of direct developers. The analyses presented here suggest that this scale is at a within shore level, although finer-scale measurements may be needed to further partition the residual variance, with the appropriate patch scale reflecting behavioural and demographic differences between different species of direct developers.

Acknowledgements

M.J. was supported by Natural Environment Research Council fellowship and grant GR3/11777. The National Museums of Scotland funded fieldwork. Port Erin Marine Laboratory assisted with facilities in the Isle of Man. T. Crowe provided useful comments on an earlier version of the manuscript.

Received 4 September 2000; revision received 27 February 2001

Ancillary